Adds modular SERDE approach

[skrawcz]

This PR introduces a modular SERDE approach.

1. Singledispatch to register type serializers.
2. Any complex object deserialization, needs to be accompanied with a "type field" that will map to the name of a "string dispatcher function". i.e. like single dispatch, but you register a string value for it. This approach means that most people will get around the need for field level deserialization / having to tell Burr about state types explicitly.

This skips field level serde registration for now. But leaves room for it.

Otherwise this custom serde for:

• langchain docs & messages
• pydantic objects
• some more complex serialization, e.g. pandas dataframe serialization
• ability to register classes to be pickled

TODOs:

• [x] decide on module structure and where it should live
• [x] wire changes through to tracker serialization
• [x] wire changes through to persister serialization
• [x] ~wire changes through to UI reading local tracker logs~ Not required.
• [x] update LC examples
• [x] write tests
• [x] add documentation

Changes

How I tested this

• unit tests
• running lcel example

Notes

Checklist

• [ ] PR has an informative and human-readable title (this will be pulled into the release notes)
• [ ] Changes are limited to a single goal (no scope creep)
• [ ] Code passed the pre-commit check & code is left cleaner/nicer than when first encountered.
• [ ] Any change in functionality is tested
• [ ] New functions are documented (with a description, list of inputs, and expected output)
• [ ] Placeholder code is flagged / future TODOs are captured in comments
• [ ] Project documentation has been updated if adding/changing functionality.

[skrawcz]

Update: just need to test more, but I think I have a good design (and update libraries to latest locally...)

core.serde has functions. core.state uses it and imports integrations state.State knows how to serialize & deserialize using it.

People can then register their own for their application -- I haven't tested overriding but I assume it'll work. Added example of pandas serializing somewhere and just showing link back...

Tracker models should be dumb -- so instead I push serialization calling into hooks. Also persisters updated to use serialization.

Once I merge and rebase off of #215 can add persister tests that will also then check for this...