fix(gateway): clear last-resolved-model cache on /new and compression auto-reset

[liuhao1024]

What does this PR do?

Clears the per-session _last_resolved_model cache when /new (or /reset) is issued, and on compression-exhausted auto-reset. This ensures the next turn resolves the model from the current config.yaml instead of falling back to a stale cached value.

Before this fix: After changing config.yaml model (e.g. deepseek-chat → mimo-v2.5-free) and running /new, the _last_resolved_model[session_key] dict still held the old model name. If the first post-/new config read returned empty (transient mtime-cache miss — the #35314 race), the recovery path served the stale cached model instead of surfacing the config-miss error.

After this fix: /new clears both _session_model_overrides AND _last_resolved_model[session_key]. The process-wide "*" fallback is preserved as a safety net for other sessions.

Related Issue

Fixes #58403

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• gateway/slash_commands.py: Added _last_resolved_model.pop(session_key, None) in _handle_reset_command after the existing _session_model_overrides.pop call. Uses getattr guard for bare test runners that skip __init__.
• gateway/run.py: Added the same _last_resolved_model.pop in the compression-exhausted auto-reset path (line ~11398), which performs the same session-model cleanup as /new.
• tests/gateway/test_new_clears_last_resolved_model.py: 3 regression tests verifying (1) per-session cache cleared on /new, (2) global "*" fallback preserved, (3) stale model not served after config change + empty config read.

How to Test

1. Run python -m pytest tests/gateway/test_new_clears_last_resolved_model.py tests/gateway/test_empty_model_recovery.py -q — all 12 tests should pass.
2. The existing test_empty_model_recovery.py tests (9 tests) verify the [Bug] model= empty on recovery turn after stream_interrupt_abort — "trying fallback..." logged but never executed, session goes silent #35314 safety net still works: transient empty config reads recover via _last_resolved_model on normal turns.
3. The new tests verify that after /new, the per-session cache is cleared so a config change takes effect on the next message.

Checklist

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/gateway/test_new_clears_last_resolved_model.py tests/gateway/test_empty_model_recovery.py -q and all tests pass
• I've added tests for my changes
• I've tested on my platform: macOS
• I've updated relevant documentation — N/A
• I've updated cli-config.yaml.example — N/A
• I've updated CONTRIBUTING.md or AGENTS.md — N/A
• I've considered cross-platform impact — N/A (pure Python logic, no platform-specific behavior)
• I've updated tool descriptions/schemas — N/A