Bug Description

When auxiliary.session_search.provider is set to auto (the default),
and the environment has no credentials for OpenRouter or Nous Portal, the
auto-detect chain still walks through them sequentially before falling
through to custom_providers. Each failed hop adds unnecessary latency,
and the subsequent fallback retry can hit the same provider again,
compounding the waste.

Steps to Reproduce

1. Configure a main provider via custom_providers (no OpenRouter or
   Nous API keys)
2. Leave auxiliary.session_search.provider: auto (default)
3. Call session_search on sessions that require LLM summarization

What Happens

The auto-detect chain walks:

auto → OpenRouter (no API key, fails) 
    → Nous Portal (no API key, fails) 
    → custom_providers (reaches target provider)
    → API call may timeout
    → fallback retries same provider
    → timeout again
    → repeats up to 3 times

From logs, a single session_search summarization call took ~560 seconds
due to this chain + retry loop. When an off-work wrapup cron reviews
multiple sessions, the cumulative delay is severe.

Expected Behavior

• If no credentials exist for OpenRouter/Nous, skip them immediately
  rather than attempting and waiting for failure
• When fallback is triggered because the API timed out, try a genuinely
  different provider instead of the same one
• Or: provide a config option to explicitly bypass the auto-detect chain
  and go directly to a named provider

Workaround

Explicitly set session_search.provider: custom:<name> in
auxiliary config to skip the auto-detect walk entirely.

Environment

• Hermes Agent: 0.8.x
• Main provider: custom_providers
• Auxiliary affected: session_search