Rework attributes to be instance based

[halgari]

Pretty massive overhaul of the attribute system to handle #37. More code was deleted here than added. And the net effect is that we're back to 400k datoms/sec insert on my machine.

What this PR does:

• Moves most of the dictionary lookups AttrId -> Attribute, AttrId -> ValueSerializer, etc. into inline arrays.
  • Since attributes are always in the range of 1 to 2^16 we can pre-allocate slots for all the attributes.
  • Likewise, when mapping attributes to DB Ids the chances of us having more than X database connections open at once is very rare, so we can pre-allocate those caches as well. X is currently set to 8.
• Transactions now work on pooled writers instead of by having each insert be a IWriteDatom class. This means that when writing data we never allocate a datom, just write the data straight to a buffer and then the transaction system streams it to disk.
  • Some improvement in the future here for constructing buffers of data during Transactor construction and never copying them until we write the transaction to storage
• Reworks all the test models to use the new format
• Reduces the number of virtual calls in a lot of places, we now use Attribute directly insteadof via IAttribute.
• Cleans up the parameters used when creating Attribute<T>, and uses enums were useful.
• Moves AttributeId from a ulong to a ushort. The values will never be higher than 2^16 anyway, but we had them as ulongs as technically they are EntityIds.
• Cleans up the word salad that was the separation of models and attributes. Now we define attributes in a static class, with the models defined as subclasses. So we can write File.Path, File.Size and read the data via File.Model