refactor(web): extract model-assignment helpers into web_server_model_assignment (web_server.py god-file slice R1-C9)

[andrexibiza]

Summary

Web_server.py god-file slice R1-C9: extract the model-assignment helpers from hermes_cli/web_server.py into hermes_cli/web_server_model_assignment.py (pure module, no app coupling). Part of the repo-wide large-file decomposition (tracker #78647).

What changed and why

• Moved _normalize_main_model_assignment + _apply_main_model_assignment byte-verbatim (window 1423–1593) into hermes_cli/web_server_model_assignment.py
• web_server.py re-exports both names (seam identity: web_server.X is web_server_model_assignment.X)
• Repointed the pre-existing test monkeypatch targets from web_server.load_config → web_server_model_assignment.load_config (function globals now live in the new module); test_moa_set_models_preserves_extra_keys.py correctly keeps patching web_server.load_config (direct call in web_server's own namespace)
• Added seam-identity regression (5) + aggressive tests (27): provider-inference edge cases, empty/invalid model strings, import isolation

Why this matters to users

web_server.py (17,700 lines at this slice's base) is one of the repo's 20 god files. Extracting cohesive clusters one PR at a time makes the dashboard surface auditable, testable, and safe to evolve — the large-file decomposition is the prerequisite for every subsequent fix on this surface.

Testing

• pytest tests/hermes_cli/test_web_server_model_assignment_seam.py tests/hermes_cli/test_web_server_model_assignment_aggressive.py tests/hermes_cli/test_normalize_main_model_assignment.py → 40 passed
• Canonical runner (scripts/run_tests.sh, hermetic env): 3 files, 36 passed, 0 failed
• Seam identity verified at runtime (is-identity, __module__ check)
• ruff clean · git diff --check clean · LF-only · DCO signed
• Platforms tested: Windows (native), CI-parity runner

Coordination / interlock

• Epic: Part of [COMPLETE] Large-file decomposition: 20/20 done #78647 (Shard all 20 god files) — this PR implements a slice of that epic's web_server.py track
• Related: the other web_server slices (gateway-topology, managed-files, schema, custom-endpoints, messaging, whatsapp) ship as sibling PRs — disjoint clusters, no overlap
• Dedup: no existing PR extracts these helpers (verified via search before implementation)
• Credit: extraction pattern follows the run.py Extract Gateway Platform Routing from gateway/run.py #54962 slices and the Telegram adapter wave (tracker Telegram Feature Parity & Alignment Campaign (Bot API 10.2) — meta-issue #78791 lessons)

Part of #78647
Part of #78628

[andrexibiza]

🔒 This shard is part of the web_server.py KILL LOCK — the permanent record of the 17,700-line whole, the mess it caused (#58576 event-loop stalls, #60800 cold-start stalls, #53972 token persistence, #71778 analytics), every shard in the wave, and every open fixer PR still fighting the surface (#50188 #71802 #54034 #76764 #71150 #67365 #65434 #65424 #78411 #75845 #68022 #62654 #76757 #68251 #51070 #73605 #73849 #73393 #71752 #77808 #46956 #74297 #77853).

Lock: posted on #78628 · Indexed by the Kill All Gods meta-issue #78647.