fix: biomodel search no longer fails for signed-in users without a linked VCell account

[androemeda]

Summary

Biomodel search returned nothing at all — not even public models — for a user who was signed in but hadn't linked a VCell account. Logged-out users were unaffected, which made it look like an auth problem when it was the opposite: having a token was what broke it.

Browsing public biomodels now works in every auth state, and results say whether they include private data.

Root cause

fetch_biomodels exchanged the Auth0 token for a legacy VCell token whenever any token was present:

VCell only issues a legacy token to a linked login. UserRestService.getUserFromIdentity throws NotAuthenticatedWebException when the caller has no UserIdentity row, so /api/v1/users/bearerToken answers an authenticated-but-unlinked caller with:

{ "exceptionType": "NotAuthenticatedWebException", "message": "User is not authenticated." }

raise_for_status() turned that into an exception that propagated out before the biomodel query was ever issued, so the controller returned 401 {"detail": "Error fetching biomodels."} and zero models. Logged-out users skipped the exchange entirely, which is why only signed-in users saw the failure.

The fix

get_legacy_vcell_token now returns Optional[str] — None when a token can't be obtained — instead of raising, and both call sites attach the header only if they got one:

Any exchange failure degrades to public-only, not just 401/403 — a 401/403 logs at info (expected: not linked), anything else logs at warning. Browsing must not depend on an auth endpoint being healthy.

fetch_biomodels also now returns includes_private in its metadata, so callers can tell "you have no private models" apart from "we couldn't check". The search page uses it to show a hint linking to /profile when you're signed in but seeing public-only results.

Behaviour after this change

State	Biomodel search & diagrams	Private / shared models	LLM features
Logged out	✅ public models	❌	❌ 401
Signed in, not linked	✅ public models + hint to link	❌	✅ works, public model data
Signed in, linked	✅ public models	✅ included	✅ works, private model data

LLM features require sign-in only, not linking. This was considered and deliberately rejected: enforcing it meant a mappedUser call on every LLM request, measured at ~1–1.5s, to guard something any user can satisfy in seconds by linking. Not worth the latency on every message.