feat: DIA-2067: Add message type counters in MessageBuilder

[niklub]

No description provided.

[Copilot]

Pull Request Overview

This PR introduces message type counters in the MessageBuilder functionality and updates tests and usage reporting accordingly. Key changes include:

• Adding "message_counts" to expected results in tests.
• Adjusting LLM usage functions to include message type counts.
• Updating dependencies and refactoring message parsing code.

Reviewed Changes

Copilot reviewed 9 out of 11 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/test_stream_inference.py	Added "message_counts" field to the expected output dictionary.
tests/test_multiimage_support.py	Updated test assertions and added message trimming test logic.
tests/test_llm.py	Updated expected cost values and "message_counts" in test outputs.
pyproject.toml	Added mypy dependency.
adala/utils/parse.py	Refactored message parsing models and removed the split function.
adala/utils/llm_utils.py	Incorporated message counters into usage data and refactored LLM calls.

Files not reviewed (2)

• .cursor/rules/common-rules.mdc: Language not supported
• .cursorignore: Language not supported

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 94.71154% with 11 lines in your changes missing coverage. Please review.

Project coverage is 70.23%. Comparing base (9bb28c2) to head (6e8a1b4).

Files with missing lines	Patch %	Lines
adala/utils/message_builder.py	94.77%	7 Missing ⚠️
adala/utils/llm_utils.py	91.66%	3 Missing ⚠️
adala/utils/token_counter.py	95.83%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #369      +/-   ##
==========================================
+ Coverage   69.07%   70.23%   +1.15%     
==========================================
  Files          49       51       +2     
  Lines        2862     2980     +118     
==========================================
+ Hits         1977     2093     +116     
- Misses        885      887       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull Request Overview

This PR implements message type counter support in the MessageBuilder and updates tests and related utilities to track message counts. Key changes include:

• Adding a "_message_counts" field to usage data in multiple test files and LLM utility functions.
• Updating tests in various modules to assert the presence and correctness of message counts.
• Removing legacy functions (e.g. split_message_into_chunks) and updating return types in MessagesBuilder.

Reviewed Changes

Copilot reviewed 13 out of 15 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_stream_inference.py	Adds a "_message_counts" entry to the expected test output
tests/test_multiimage_support.py	Adds a new test to validate message trimming and confirms message counts
tests/test_llm_utils.py	Adds tests for message type counting
tests/test_llm.py	Updates expected test outputs; removes split_message_into_chunks test
tests/conftest.py	Updates VCR configuration to filter additional query parameters
pyproject.toml	Upgrades pytest and adds mypy for code quality checks
adala/utils/token_counter.py	Introduces a new token counter implementation using LiteLLM
adala/utils/parse.py	Removes legacy message chunk splitting code, streamlining template parsing
adala/utils/llm_utils.py	Modifies usage dict creation to include message counts and adjusts message extraction from MessagesBuilder

Files not reviewed (2)

• .cursor/rules/common-rules.mdc: Language not supported
• .cursorignore: Language not supported

Comments suppressed due to low confidence (4)

tests/test_llm.py:148

• Removal of the test_split_message_into_chunks function may reduce test coverage for the message chunk splitting functionality. Consider retaining or replacing these tests if the functionality is still required.

def test_split_message_into_chunks():

adala/utils/llm_utils.py:351

• The usage of messages_builder.get_messages(payload).messages assumes that get_messages now returns an object with a 'messages' attribute. Ensure that this change is consistent throughout the codebase and that the return type is properly documented.

messages = messages_builder.get_messages(payload).messages

adala/utils/llm_utils.py:420

• The change to access a 'messages' attribute from get_messages(payload) needs consistency checks across all consumers of this method. Please verify that all related code and documentation are updated accordingly.

messages = messages_builder.get_messages(payload).messages

adala/utils/llm_utils.py:492

• Ensure that returning an object with a 'messages' property from get_messages(payload) is the intended design, and that all invocations are updated consistently to prevent potential runtime errors.

messages = messages_builder.get_messages(payload).messages

[matt-bernstein]

Please no abstract classes here :( keep it to the minimum layers of abstraction necessary, here you can just use the litellm token count functions directly https://docs.litellm.ai/docs/completion/token_usage

[niklub]

We're trying to build the codebase using SOLID principles to resolve potential dependency inversion problems. Specifically, every time you design a new token system, a method of tokenizing, or rely on different ways of counting tokens, you end up with a tightly coupled dependency on litellm. I envision a more sophisticated token management scenario in the future; therefore, decoupling and abstracting this component is essential, as well as in general following the standards of code development.

[matt-bernstein]

can you remove cursorrules from this PR so we can review it separately?