Streaming chunk accumulation

[nichwch]

Implements new validate_stream method on Validator, and changes stream_runner to pass individual chunks to validators. Validators now internally accumulate chunks until they reach a specified chunk size, then emit a validation up to the iterator in stream_runner.

These changes only apply to string_schema and SequentialValidatorService for now.

EDIT: This now contains the changes for the error_spans schema changes as well

[zsimjee]

1: In stream_runner, how can I tell what the validator chunk size is? I need to know this to decide if the validator chunk size is smaller than the LLM chunk size, in which case I'll have to split the validator chunk size to be smaller.

See comments in stream_runner, string_schema

2: Is it true that the parse method for string_schema is just supposed to check that the output isn't empty?

It also makes sure that the output is a string. i.e. not num, obj, etc...

3: What would a good way to test these changes be? What general workflow do you guys use to test your changes (besides the built in tests)

best way to check for backwards compatability would be by using the integration tests + running the streaming noteboook (https://github.com/guardrails-ai/guardrails/blob/14bb0bccde051338dbc29cc50c77970a4bfb1a08/docs/how_to_guides/streaming.ipynb)

[zsimjee]

Good to approve this pending test and checks.