fix(google-genai): include cached_content_token_count in streaming responses

[Aftabbs]

Summary

Fixes #3168

cached_content_token_count from the Google GenAI API usage metadata was
being populated in non-streaming responses but was completely absent
from streaming responses.

Root cause (two missing pieces):

1. _convert_google_chunk_to_streaming_chunk only checked for
   thoughts_token_count but had no equivalent check for
   cached_content_token_count. The usage dict built for each chunk
   therefore never included the cached token count.

2. _aggregate_streaming_chunks_with_reasoning forwarded
   thoughts_token_count from the final chunk into the aggregated
   message, but had no corresponding logic for
   cached_content_token_count.

Fix:

• Add an explicit cached_content_token_count check in
  _convert_google_chunk_to_streaming_chunk, matching the pattern
  already used by the non-streaming path.
• Propagate cached_content_token_count through
  _aggregate_streaming_chunks_with_reasoning, mirroring how
  thoughts_token_count is handled.
• Add two unit tests covering the streaming chunk conversion and the
  aggregator.

How to test

from haystack.dataclasses import StreamingChunk, ComponentInfo
from haystack_integrations.components.generators.google_genai.chat.utils import (
    _convert_google_chunk_to_streaming_chunk,
    _aggregate_streaming_chunks_with_reasoning,
)
from unittest.mock import Mock

# Streaming chunk with cached tokens
mock_usage = Mock()
mock_usage.prompt_token_count = 1000
mock_usage.candidates_token_count = 10
mock_usage.total_token_count = 1010
mock_usage.thoughts_token_count = None
mock_usage.cached_content_token_count = 800
# ... (build mock chunk) ...
assert chunk.meta["usage"]["cached_content_token_count"] == 800

# Aggregated message
result = _aggregate_streaming_chunks_with_reasoning([chunk1, final_chunk])
assert result.meta["usage"]["cached_content_token_count"] == 800

[Copilot]

Pull request overview

Fixes missing cached_content_token_count in streaming responses for GoogleGenAIChatGenerator by extracting it from chunk usage metadata and propagating it into the aggregated ChatMessage usage metadata.

Changes:

• Add cached_content_token_count extraction in _convert_google_chunk_to_streaming_chunk.
• Propagate cached_content_token_count from streaming chunks into _aggregate_streaming_chunks_with_reasoning.
• Add unit tests covering both the per-chunk conversion and final aggregation behavior.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
integrations/google_genai/src/haystack_integrations/components/generators/google_genai/chat/utils.py	Extract and forward cached_content_token_count in streaming chunk conversion and aggregation.
integrations/google_genai/tests/test_chat_generator_utils.py	Add tests ensuring cached token count appears in streaming chunk meta.usage and aggregated message meta.usage.
integrations/google_genai/CHANGELOG.md	Document the bug fix under [Unreleased].

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The condition and usage_metadata.cached_content_token_count treats 0 as absent, so cached_content_token_count=0 would be dropped even if the API explicitly returns it. Use an is not None check (e.g., assign via getattr(..., None) and compare to None) to preserve valid zero values and match the intent of “if available”.

Suggested change

	if (
	usage_metadata
	and hasattr(usage_metadata, "cached_content_token_count")
	and usage_metadata.cached_content_token_count
	):
	usage["cached_content_token_count"] = usage_metadata.cached_content_token_count
	cached_content_token_count = getattr(usage_metadata, "cached_content_token_count", None) if usage_metadata else None
	if cached_content_token_count is not None:
	usage["cached_content_token_count"] = cached_content_token_count

[Copilot]

Call _convert_google_chunk_to_streaming_chunk using named keyword arguments (as done elsewhere in this test module) to keep the test consistent and resilient to parameter reordering.

Suggested change

	chunk = _convert_google_chunk_to_streaming_chunk(mock_chunk, 0, component_info, "gemini-2.5-flash")
	chunk = _convert_google_chunk_to_streaming_chunk(
	chunk=mock_chunk,
	candidate_index=0,
	component_info=component_info,
	model="gemini-2.5-flash",
	)

[github-actions]

Coverage report (google_genai)

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
integrations/google_genai/src/haystack_integrations/components/generators/google_genai/chat
utils.py					550, 748, 754
Project Total

This report was generated by python-coverage-comment-action

[bogdankostic]

Thanks @Aftabbs, looks good overall, just left two minor comments.

[bogdankostic]

Please revert the changes made here, the Changelog is generated automatically whenever we release a new version.

[bogdankostic]

Let's update this comment, we don't only extract thinking token usage here.

[bogdankostic]

Looking good to me, thanks @Aftabbs!