fix(compression): 3-line fix for infinite compression loop (#29335)

[someaka]

Problem

When context compression fires, three compounding issues cause an infinite compression loop:

Issue A — session_id not propagated

_compress_context() updates self.session_id to the new session, but run_conversation() never includes session_id in its return dict. The gateway at gateway/run.py:8700 has a guard but it's dead code — the agent never sends session_id.

Issue B — preflight bypasses anti-thrashing

Preflight compression uses a raw threshold comparison, bypassing should_compress() which has anti-thrashing logic. When system prompt + tool schemas dominate token count, compressing messages doesn't bring total below threshold — preflight re-triggers every turn despite saving <10%.

Issue C — gateway doesn't persist session_id update

Gateway updates session_entry.session_id in memory but doesn't call session_store._save(). Two other compression paths in the same file DO persist — this one was missed.

Fix — 4 lines, 2 files

 agent/conversation_loop.py | 3 +++
 gateway/run.py              | 1 +
 2 files changed, 4 insertions(+), 1 deletion(-)

Fixes: #29335

[alt-glitch]

Competing with open PRs #29505 (3-part fix: should_compress() + session_id + _save()), #26204, and #26088 — all fix #29335 (infinite compression loop due to session_id not propagated after compression). Clean replacement of stacked #31725. This PR addresses issues A (session_id not in result dict), B (preflight bypasses anti-thrashing), and C (gateway doesn't persist session_id update).

[teknium1]

Merged via #31930 — your commit was cherry-picked onto current main with authorship preserved (3914089d5). Thanks for the fix and the clean root-cause writeup!

Follow-up commit (11c40d6a4) on top:

• Polish: dropped the redundant _preflight_tokens >= threshold_tokens clause next to your new should_compress(_preflight_tokens) call (should_compress already short-circuits below threshold, so the explicit comparison was dead code on the True branch).
• Tests: pinned the anti-thrash gating (mocks should_compress to return False even with tokens past threshold and asserts no compression runs — exact bug shape from Bug: Context compression creates new session but gateway never sees it — infinite compression loop #29335) and added an AST-scan test that asserts every session_entry.session_id = ... assignment in gateway/run.py is followed by a session_store._save() call within the same block. Empirically verified the second test catches the bug by reverting the new _save() line → red, restoring → green.
• AUTHOR_MAP entry mapping ed@bebop.crew to @someaka for the release notes.

Closes #29335.