fix(cron): fall back to auto-detect provider when explicit provider fails

[GratefulDave]

Problem

When a cron job's configured provider (from config.yaml model.provider or HERMES_INFERENCE_PROVIDER env) has no valid credentials — e.g. the config references anthropic but ANTHROPIC_API_KEY is not set — the job fails immediately with an AuthError. The entire cron run is lost.

This is especially painful after initial setup where config.yaml may contain a default or stale provider that doesn't match the available credentials.

Fix

When resolve_runtime_provider() raises for the explicitly configured provider, catch the exception and retry with requested="auto" to find any working provider before giving up.

The original error is logged as a warning, so operators can see that a fallback occurred and fix the config:

Job my-job: provider resolution failed (No Anthropic credentials found), falling back to auto-detect

If auto-detection also fails, the original hard error is preserved.

Testing

• Cron job with config.yaml pointing to anthropic but no Anthropic key: previously failed, now falls back to auto-detecting a working provider
• Cron job with valid explicit provider: no change in behavior
• Cron job with no configured provider at all: already uses auto, no change

[GratefulDave]

Closing — needs a fresh rebase against current main. Will resubmit if still relevant.