fix: preserve CrewAI tool history across turns

[darvell-thenvoi]

Summary

• preserve CrewAI tool_call/tool_result events as replayable history instead of dropping them
• replay the full room transcript on each CrewAI kickoff instead of reducing prior turns to a bootstrap blob
• add regression coverage for tool replay and full-history kickoff behavior, and update CrewAI converter conformance expectations
• change platform instructions order so platform instructions are the first thing rather than appended

Test plan

• env -u VIRTUAL_ENV uv run python -m pytest tests/converters/test_crewai.py tests/adapters/test_crewai_adapter.py -v
• env -u VIRTUAL_ENV uv run python -m pytest tests/framework_conformance/test_converter_conformance.py -v -k crewai
• env -u VIRTUAL_ENV uv run ruff check src/thenvoi/converters/crewai.py src/thenvoi/adapters/crewai.py tests/converters/test_crewai.py tests/adapters/test_crewai_adapter.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py
• env -u VIRTUAL_ENV uv run ruff format --check src/thenvoi/converters/crewai.py src/thenvoi/adapters/crewai.py tests/converters/test_crewai.py tests/adapters/test_crewai_adapter.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py
• env -u VIRTUAL_ENV uv run pyrefly check src/thenvoi/converters/crewai.py src/thenvoi/adapters/crewai.py

[darvell-thenvoi]

Rechecked current head. No new code changes were needed; the PR remains scoped to CrewAI history/tool replay handling and its converter/conformance coverage.

Validation:

• uv run --extra dev pytest tests/adapters/test_crewai_adapter.py -k 'history or replay or backstory or tool' -q --no-cov -> 45 passed, 25 deselected
• uv run --extra dev pytest tests/converters/test_crewai.py tests/framework_conformance/test_converter_conformance.py -k crewai -q --no-cov -> 28 passed, 1 skipped, 154 deselected
• uv run ruff check src/thenvoi/adapters/crewai.py src/thenvoi/converters/crewai.py tests/adapters/test_crewai_adapter.py tests/converters/test_crewai.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py -> passed
• uv run ruff format --check src/thenvoi/adapters/crewai.py src/thenvoi/converters/crewai.py tests/adapters/test_crewai_adapter.py tests/converters/test_crewai.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py -> passed

[darvell-thenvoi]

Rechecked current head. No new code changes were needed; the PR remains scoped to CrewAI history/tool replay handling and its converter/conformance coverage.

Validation:

• uv run --extra dev pytest tests/adapters/test_crewai_adapter.py -k 'history or replay or backstory or tool' -q --no-cov -> 45 passed, 25 deselected
• uv run --extra dev pytest tests/converters/test_crewai.py tests/framework_conformance/test_converter_conformance.py -k crewai -q --no-cov -> 28 passed, 1 skipped, 154 deselected
• uv run ruff check src/thenvoi/adapters/crewai.py src/thenvoi/converters/crewai.py tests/adapters/test_crewai_adapter.py tests/converters/test_crewai.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py -> passed
• uv run ruff format --check src/thenvoi/adapters/crewai.py src/thenvoi/converters/crewai.py tests/adapters/test_crewai_adapter.py tests/converters/test_crewai.py tests/framework_configs/converters.py tests/framework_conformance/test_converter_conformance.py -> passed