fix(gateway): keep native images on resolved session key

[tt-a1i]

Summary

• pass the resolved gateway session key into inbound message preprocessing
• buffer native image paths under the same key consumed by _run_agent
• add a regression test for source-derived vs canonical session-key mismatch

Fixes #48912

Validation

• .venv/bin/python -m pytest tests/gateway/test_native_image_buffer_isolation.py -q -o addopts=
• .venv/bin/python -m pytest tests/gateway/test_native_image_buffer_isolation.py tests/gateway/test_video_context_note.py tests/gateway/test_shared_group_sender_prefix.py tests/gateway/test_reply_to_injection.py tests/gateway/test_stt_config.py tests/gateway/test_session.py tests/gateway/test_telegram_audio_vs_voice.py -q -o addopts=
• .venv/bin/python -m py_compile gateway/run.py
• git diff --check

Review

• Subagent read-only review: no findings; safe to commit/PR.

[alt-glitch]

Related: #48919 (competing fix for the same issue #48912, different mechanism — it re-resolves the session key for the queued follow-up re-entry in _notify_long_running, whereas this PR threads the caller-resolved session_key into _prepare_inbound_message_text at the buffer-write site). Both build on the per-session native-image-buffer isolation introduced by merged #18202. Same goal, distinct code sites — not a duplicate; maintainers should pick the preferred mechanism (or combine both, since #48919 covers the queued-followup branch and this covers the direct write site).

[alaamohanad169-ship-it]

Acknowledged: noted competing PR(s) #48919, #48912, #18202, #48919. My approach differs and is complementary, not mutually exclusive. Open to consolidating if reviewers prefer one direction.

[teknium1]

Merged via PR #57648. Your commit was cherry-picked onto current main with your authorship preserved in git log (rebase-merge). We added a small follow-up on top reordering the queued-follow-up session-key resolution so it happens before the inbound-text prep buffers images. Thanks for the fix!