fix(server): surface producer errors after failed tasks

[dawNotPoi]

Description

Blocking message/send previously stopped consuming as soon as a FAILED Task replacement arrived. If the producer then raised, the subscriber never observed that exception and the request returned the failed Task as a successful RPC result.

This keeps a FAILED Task as the fallback response but continues consuming until the current request completes, the stream closes, or the producer error arrives. A clean agent-declared failure still returns its Task; a producer crash now surfaces its original exception. Other terminal states and return_immediately retain their existing behavior.

• Follow the CONTRIBUTING Guide.
• Make the Pull Request title follow the Conventional Commits specification.
• Ensure the tests and linter pass.
• Appropriate docs were updated (not necessary for this internal behavior fix).

Test plan

• ./scripts/lint.sh
• uv run pytest — 1,987 passed, 90 skipped, 3 xfailed, 1 xpassed
• uv run pytest --cov=src --cov-report=term-missing — 93% total coverage
• Focused request-handler regression matrix — 7 passed

Fixes #1228 🦕

AI assistance

Codex was used for implementation, testing, and independent code review. This Draft remains subject to human review.

[github-actions]

🧪 Code Coverage (vs main)

⬇️ Download Full Report

	Base	PR	Delta
src/a2a/server/request_handlers/default_request_handler_v2.py	92.44%	92.53%	🟢 +0.09%
Total	93.05%	93.06%	⚪️ 0.00%

Generated by coverage-comment.yml

[mykytanetipa]

LGTM