fix: include tools in post-compression token estimate

[LeonSGP43]

Summary

• use estimate_request_tokens_rough(...) when updating last_prompt_tokens after context compression
• include tool schema tokens in the post-compression estimate so the next compression cycle is scheduled against the real request size
• add a regression test covering _compress_context() with a large tool schema

Closes #14695

Testing

• python3 -m pytest -o addopts= tests/run_agent/test_context_token_tracking.py

[teknium1]

Thanks for the focused diagnosis and regression coverage. This is now implemented on main.

• Automated hermes-sweeper review verified that fix(compression): include system prompt + tool schemas in token estimates #18265 merged as f0dc919f92c5327cf8033e06c039126f1288e89c and closed BUG: Post-compression token estimate excludes tools schema, delaying next compression cycle #14695 with the same tool-inclusive estimate_request_tokens_rough(..., tools=...) fix.
• Current main keeps that estimate in agent/conversation_compression.py:969-974 and uses tool-aware preflight estimates in agent/turn_context.py:374-378 and :458-461.
• A later follow-up (e38b0b55d) deliberately changed last_prompt_tokens to await provider-reported usage after compression, preventing schema-heavy rough estimates from triggering repeated compaction.

Closing as implemented on main.