Add author and token count sorting

[Archkr]

Summary

Add Author and Token Count sorting to the Characters browser and Token Count
sorting to the Personas browser. Character author ordering is case-insensitive,
keeps blank authors last, and uses character name/id tie-breakers. Character
token totals count Description, Personality, and Scenario separately before
server-side pagination. Persona token totals count Description through a new
authenticated, user-scoped POST /api/v1/personas/token-counts endpoint.

Both token sorts resolve the active profile model, fall back to the default
profile model, and ultimately use the existing ceil(text.length / 4) behavior
when no exact tokenizer is available. Persona requests are made only while
Token Count is selected and use a generation guard so stale model/persona
responses cannot replace current counts. Group Chats continue to expose only
Recent, Name, and Created. Labels were added to all six existing locales.

No main-menu sorting button, persona author field, database migration, or schema
change is included.

Validation

Commands and results:

bun test src/services/characters.service.test.ts src/routes/personas.routes.test.ts frontend/src/lib/characterSort.test.ts frontend/src/lib/personaTokenSort.test.ts src/macros/MacroEnv.test.ts
# 32 pass, 0 fail, 64 assertions across 5 files

bun x tsc --noEmit
# Passed with no output

cd frontend && bun run typecheck
# Passed; TypeScript project build completed

cd frontend && bun run lint
# Passed with 0 errors and 0 warnings

cd frontend && bun run build
# Passed; production bundle and PWA service worker built successfully

Manual validation used an isolated temporary Lumiverse instance and disposable
30-character/30-persona data set. Verified:

• Character Author and Token Count options, both directions, blank authors last,
  stable pagination, search interaction, and persistence after reload.
• Character Token Count ordering using the fallback tokenizer path.
• Group Chats exposes only Name, Recent, and Created.
• Persona Token Count in both directions, pagination, search/filter interaction,
  and persistence after reload.
• Desktop layout and a 390x844 responsive mobile viewport.
• No warning/error entries in the validation browser console.

The isolated server and disposable validation data were removed afterward.

Required checklist

• This pull request is based on the latest staging branch and targets
  staging, not main.
  Current local branch is staging, but no remote sync or PR target was
  performed/verified.
• All applicable tests pass.
• I ran the relevant type check(s) with no type errors or warnings:
  • Backend: bun x tsc --noEmit
  • Frontend: cd frontend && bun run typecheck, bun run lint

UI changes (if applicable)

• I validated this change in more than one browser.
• I verified the integrated experience on a mobile interface or through
  the mobile PWA.
• Browsers, devices, and PWA coverage: Codex in-app browser at its desktop
  viewport and at a 390x844 responsive mobile viewport. An installed PWA and a
  second browser were not tested.

Performance changes (if applicable)

• I added or updated tests.
• I included reproducible benchmarks and comparison results below.
• Benchmark commands and results: Not run. Character token sorting intentionally
  resolves ordering before pagination and reuses the tokenizer service's loaded
  tokenizer and content-hash count caches. Focused tests cover stable pagination.

Security-sensitive changes (if applicable)

The new persona token-count endpoint uses the authenticated user id and queries
only that user's persona ids/descriptions. It accepts only an optional string
model_id. No Spindle surface, privilege boundary, database schema, or stored
credential behavior changed; no separate security audit was performed.

LLM-assisted work (if applicable)

• I, as the human orchestrating this work, audited the submitted changes
  in a live environment.