test(pydantic_ai): Remove usage of internal functions from tests

[DawnofGenX]

Summary

Refactors tests/integrations/pydantic_ai/test_pydantic_ai.py so it no longer imports or calls internal (_-prefixed) SDK functions, per CONTRIBUTING ("Don't call private functions") and the review note in #5629. Fixes #5637.

What changed (test-only; no sentry_sdk/ changes):

• Removed imports/calls of _set_input_messages, _set_usage_data, _set_output_data, _set_available_tools, _get_model_name from the test module. Tests that exercised those helpers now drive the public integration surface: ai_client_span() / update_ai_client_span() (with real pydantic_ai message/part objects — UserPromptPart, SystemPromptPart, ModelResponse, RequestUsage — instead of MagicMock messages), and read back the recorded data from the emitted chat span.
• Several of these tests were previously vacuous (_set_input_messages(span, msgs) + assert transaction is not None — no assertion on captured data). They now assert the actual span attributes (gen_ai.request.messages, gen_ai.system.instructions, gen_ai.request.model, gen_ai.usage.*, gen_ai.request.available_tools, absence of messages when include_prompts=False/PII off), so coverage is strictly increased.
• Agent-context tests: sentry_init-scoped assertions now use the integration's public helpers get_current_agent()/push_agent()/pop_agent() from sentry_sdk.integrations.pydantic_ai.utils instead of poking scope._contexts["pydantic_ai_agent"]. Note the SDK no longer stores agent state in Scope._contexts at all (it is a private contextvar now), so every one of those _contexts assertions was vacuously true; they now verify the real mechanism. The two tests that injected fake streaming/agent state into scope._contexts were rewritten to exercise ai_client_span() through push_agent()/streaming runs instead of injecting into a private dict.
• BLOB_DATA_SUBSTITUTE is now asserted as the literal wire value ("[Blob substitute]") with a comment, instead of importing the internal constant from sentry_sdk._types (removes the last from sentry_sdk._types import ... in the file).
• Renamed two tests whose names referenced internal functions (test_get_model_name_* → test_model_name_*, test_set_usage_data_with_cache_tokens → test_usage_data_with_cache_tokens).

Test plan

Full integration suite green on both supported pydantic-ai pins after refactor:

$ TESTPATH=tests/integrations/pydantic_ai uv run tox -e py3.14-pydantic_ai-latest
====================== 331 passed, 69 warnings in 12.57s =======================
  py3.14-pydantic_ai-latest: OK

$ TESTPATH=tests/integrations/pydantic_ai uv run tox -e py3.10-pydantic_ai-v1.0.18
====================== 331 passed, 68 warnings in 35.07s =======================
  py3.10-pydantic_ai-v1.0.18: OK

No private-underscore usages remain (grep is empty; exit 1 = no match):

$ grep -nE "_set_|_should_send_prompts|_get_model_name|_get_context|_get_tool_def|_contexts|from sentry_sdk\._|import _" tests/integrations/pydantic_ai/test_pydantic_ai.py
GREP_EXIT=1

Lint:

$ uv run ruff check tests
All checks passed!
$ uv run ruff format tests
<file> reformatted (only)

Fixes #5637

[alexander-alderman-webb]

Going to close this because we don't have capacity to review the 5 PRs you've opened.
If you'd like to contribute dedicate your time towards making one good PR. Thanks.