fix(p2p): do not cache lookup timeouts as misses - #152

Summary

• avoid inserting a negative descriptor-cache entry when a P2P lookup times out
• allow the next layer read to retry peer lookup instead of falling back to the origin for the full miss TTL
• add a regression test covering repeated reads after lookup timeouts

Why

A lookup timeout means the artifact availability is unknown, not that the
artifact is absent. Caching the timeout as a miss causes subsequent reads for
that layer to fall back to the origin registry until the negative-cache entry
expires.

Validation

• make test
• result: passed, exit code 0

Refs #138