feat(web): publish reputation APIs and integrate planner constraints

[dominiccreates]

Closes #307
This PR exposes bounded reputation reads and optional planner policies directly in our marketplace routing layers (GET /api/talos and GET /api/services). It natively allows clients to filter for agents and services that meet minimum score thresholds and confidence levels, while optionally bypassing bounds for "cold-start" providers.

Scope & Changes:

In-Memory Filtering with Pagination: Rather than attempting complex reputation calculations inside the SQL database layer, this PR introduces fetchReputations (a bulk-fetching utility inside src/lib/reputation-ledger.ts) to calculate reputations natively in memory using the exported MAX_JOB_AGE_DAYS sliding window.
Endpoint Upgrades (/api/talos, /api/services): The pagination loop was refactored. We now fetch pages progressively, evaluating reputations on each chunk. The endpoints respect three new policies:
minScore: Required minimum reputation (0-100)
minConfidence: Required minimum confidence bounds (0.0-1.0)
allowColdStart: Whether providers without sufficient jobs evidence pass despite not having computed confidence limits.
OpenAPI Integration: Defined #components/parameters/minScoreParam, minConfidenceParam, and allowColdStartParam mapping these planner rules directly to the OpenAPI spec schema for auto-generation.
Acceptance criteria validation:

✔️ Behavior mapped identically across both primary search endpoints.
✔️ Bounds evaluated against strict, testable functional components.
✔️ All unit tests and openapi:snapshot tests green locally.

[vercel]

@dominiccreates is attempting to deploy a commit to the Cankat's projects Team on Vercel.

A member of the Team first needs to authorize it.