* Initialize `removeCover` as `false` on each dialog render, so previous
  overrides don't affect the initial state.
* Make both file upload and cover online search update `removeCover` to
  false.
* Fix collection update to send `remove_cover` to the API.

Potential fix for #1169.

fix: Correctly update state for cover deletion

If a game is manually unmatched, we update its `name` field to
`file_name` (which includes both tags and extension). After that, if we
go to the Rom list, and click on "Manual match" for that rom, the search
field is populated with that Rom name, which commonly won't retrieve any
results from the metadata providers.

This change makes the "Manual match" search term to use the filename
without tags or extension, when the Rom is currently not matched. It
also triggers the search automatically when a search term was set.

Potential fix for #1167.

…unmatched-rom

fix: Use filename without tags or extension when matching unmatched game

[ROMM-1181] Skip hashing games on desktop platorms

[ROMM-1176] Fetch video ID and place in carousel

[ROMM-829] Status fields on rom user

Bypass nginx proxy buffering when uploading roms

[ROMM-1114] Fetch and use age ratings in UI and filters

This change replaces the bundled Redis server with Valkey. No breaking
changes are introduced, as considered environment variables still
maintain the `REDIS_` prefix.

Fixes #925.

…-of-redis

misc: Use bundled Valkey instead of Redis server

Build and include the `RAHasher` binary in the Docker image, to
calculate hashes for RetroAchievements. Also, add a service to
run `RAHasher` from Python.

Example usage:

```python
from adapters.services.rahasher import RAHasherError, RAHasherService

rahasher = RAHasherService()
try:
    hash = await rahasher.calculate_hash("nes", Path("path/to/rom.nes"))
except RAHasherError:
    # Handle error
    hash = None
```

feat: Add RAHasher and util to calculate RetroAchievements hashes

When serving files using the `X-Accel-Redirect` header in nginx, the
header values must be URL-encoded. Otherwise, nginx will not be able
to serve the files if they contain special characters.

This commit adds a new `FileRedirectResponse` class to the `utils.nginx`
module, to simplify the creation of responses that serve files using the
`X-Accel-Redirect` header.

Fixes #1212, #1223.

…erving-nginx

fix: Escape URLs when files are served by nginx

This change improves memory usage, by only keeping a single archive's
member file in memory at a time during 7zip decompression.

The `py7zr` library does not support streaming decompression yet, so
this change is the best we can do for now.

Potential fix for #1211, but it won't improve memory usage for
single-file 7zip archives.

…ompression

fix: Improve memory usage during 7zip decompression

According to Ruffle documentation [1], the `preferredRenderer` option is
only needed if a renderer needs to be enforced. When not provided:

> By default, Ruffle chooses the most featureful backend supported by the
> user's system, falling back to more basic backends if necessary.

This commit removes the `preferredRenderer` option from the Ruffle
configuration in the frontend, so that the default behavior is used.

It also removes the noisy `logLevel` option, which was set to `debug`.

Fixes rendering issue for `Boxhead: The Rooms`, reported in #1216.

[1] https://github.com/ruffle-rs/ruffle/wiki/Using-Ruffle

…ers-renderer

fix: Do not set Ruffle's renderer explicitly

The `scan.py` module was moved to the `sockets` package, but this line
was not updated to reflect that change. This currently causes the
watcher to not skip queuing a new `scan_platforms` task if one is
already scheduled.

This change replaces the creation of Redis URL, from a simple string
interpolation, to using `yarl.URL`. The main benefit, besides not
forgetting to set all five different variables on every Redis client
initialization, is that user credentials are correctly URL-encoded, if
present.

Up until now, if a password had special characters, it could break the
generated URL.

This change also introduces support for a `REDIS_SSL` setting, which
allows the user to specify if the Redis connection should use SSL or not.

The previous implementation was calling `resize_cover_to_small` within
the context manager that was writing the image to the filesystem. This
was causing `PIL` to raise an error because it could not identify the
open and temporarily created file as a valid image.

Instead of saving the original image to the filesystem and then resizing
it, we now open the image in memory, resize it, and then save it to the
filesystem. We also avoid reading the `BytesIO` object twice by saving
small and big images from the same initial `Image` object.

Fixes #1191.

when a platform slog is not present in `PLATFORMS_VERSIONS`, which is
the common case overall, the component was trying to fetch an
`undefined.ico` image.

Instead of that, we should check if the platform slug is present in
`PLATFORMS_VERSIONS` and if not, fallback to the platform slug itself.

fix: Improve handling of Redis URL

…heduled

fix: Skip scanning task by checking against the correct function name

fix: Correctly resize and save small artwork

…efined.ico

fix: Avoid image requests for undefined.ico platform icon

misc: Upgrade RALibretro to v1.8.0

The `add_user` endpoint was querying the database at import time, to
decide whether to enforce the `users.write` scope or not. This is
problematic because the database might not be ready at import time.

Also, the decided `scopes` was being maintained for the entire
application lifetime, which is not ideal, as users can be created
without having the `users.write` scope, until the application is
restarted.

Add offset parameter to get_roms endpoint

…ort-time

fix: Avoid query during import time to set add_user endpoint scopes

Instead of fetching all users and checking if the new username is present
in the list, we can directly query the database for the username.

Switched to MARIADB variables as we are using the MARIADB container

Added comment directing people. to the example docker compose file. Changed variables to MARIADB as we are using the MARIADB container

Update variables to MariaDB

…-timeout

#1228 Configurable upload timeout

…username-exists

fix: Simplify query that validates new username already exists

Instead of using just strings, this change converts the scopes to a
`StrEnum`, to be compatible with places where a string is expected. This
avoids typos when using these scopes, simplifies searching for usages,
and improves type hints.

An extra change was the fix to the Firmware download endpoint, which
wasn't respecting the `DISABLE_DOWNLOAD_ENDPOINT_AUTH` flag.

misc: Create enum for authorization scopes

docs: Updated readme preview

…ig-exclusions-from-UI

…s-from-UI

feat: Enable editable exclusions from webUI

The `tar` decompression function was failing for some users, with error
message:

```
'NoneType' object does not support the context manager protocol
```

As explained in the official documentation [1], the `extractfile` method
returns `None` if the member is not a regular file or a link. This
change skips any member that is not a regular file.

[1] https://docs.python.org/3/library/tarfile.html#tarfile.TarFile.extractfile

Some of the image links for IGDB age ratings were broken. This commit
fixes all of them.

The IGDB API provides screenshots in a 1280x720 resolution, when either
`screenshot_huge` or `720p` size values are used [1]. The
`screenshot_huge` uses `Lfill`, while `720p` uses `Fit` as the resizing
method.

That means that either `screenshot_huge` or `720p` will provide the
correct aspect ratio for 16:9 platforms. However, for platforms with a
different aspect ratio, we need to use the `720p` size value to get the
correct image instead of a cropped one.

Example:
- [`screenshot_huge`](https://images.igdb.com/igdb/image/upload/t_screenshot_huge/c77bodnkwu73gs2jwhw4.jpg)
- [`720p`](https://images.igdb.com/igdb/image/upload/t_720p/c77bodnkwu73gs2jwhw4.jpg)

[1] https://api-docs.igdb.com/#images