[SYNPY-1544] Synapse Agent OOP Model

[BWMac]

Description:

This PR is the implementation of an OOP model for interacting with the new Synapse Agent API endpoints. The design for the model implementation can be found in Confluence.

The core implementation includes (from "bottom to top"):

• synapseclient.api.agent_services:
  • Low-level functions for interacting with the API endpoints that are specific to using Synapse Agents.
• synapseclient.models.AgentPrompt:
  • A dataclass representing a prompt sent to a Synapse Agent within a chat session.
  • Holds prompt, response and all necessary metadata for tracking an individual prompt.
  • Only has an async interface for its primary business logic - this class is used by the higher-level AgentSession class and is not intended to be used directly.
  • The concept of a "prompt" in the Synapse API is built on the more general concept of an "asynchronous job" which is also used for operations such as large data transfers and the minting of DOIs.
    • https://rest-docs.synapse.org/rest/org/sagebionetworks/repo/model/asynch/AsynchronousResponseBody.html
    • https://rest-docs.synapse.org/rest/org/sagebionetworks/repo/model/asynch/AsynchronousRequestBody.html
    • https://rest-docs.synapse.org/rest/org/sagebionetworks/repo/model/asynch/AsynchronousJobStatus.html
    • Previously, we did not have a generalized solution for creating OOP models on top of the asynchronous job functionality, and so we (@BryanFauble helped out quite a bit here) implemented the AsynchronousCommunicator Mixin which includes reusable logic for starting an async job, waiting for it to complete, and returning the response.
• synapseclient.models.AgentSession:
  • a dataclass representing a chat session with a Synapse Agent.
  • Includes functionality for directly interacting with chat sessions with a Synapse Agent.
  • The AgentSession logic is used in the Agent class, and we will generally recommend that users utilize the Agent class as the primary interface.
  • Asynchronous and Synchronous interfaces are available.
• synapseclient.models.Agent:
  • a dataclass representing a Synapse Agent you can interact with.
  • Includes functionality for registering a custom agent, starting sessions with agents, sending prompts and also managing more than one session at a time.
  • The basic use case includes one session at a time where a session started by a user (either explicitly with start_session or implicitly by just calling prompt) becomes the default for prompting, but examples for conducting more than one session at a time can be found in the POC script and class docstring.

Notes:

• All needed unit and integration tests for synchronous and asynchronous implementations are included.
• Example usage of the models can be found in the class docstrings and in the POC script included in the PR.
• I took the opportunity to refactor the Experimental section of the documentation, breaking up each model into it's own markdown script which makes maintaining the docs for OOP models a little easier.

[BryanFauble]

@BWMac A bunch of tests seem to be failing with:

FAILED tests/integration/synapseclient/models/async/test_agent_async.py::TestAgentPrompt::test_send_job_and_wait_async_with_post_exchange_args - synapseclient.core.exceptions.SynapseHTTPError: 400 Client Error: AgentRegistrationId='29' does not exist

Since the integration tests run against the DEV Instance of Synapse, does this agent need to be created there, and tests updated with this config?

[BWMac]

@BryanFauble Ah yes the agent will need to be registered in Synapse Dev.

[BryanFauble]

Awesome changes! This all looks good to me.

When all the tests are passing we can move forward on merging the changes.