Uncapped Indexing & Split Provider Timeouts

Agents Remember 0.9.1

Splits the overloaded timeoutCaps.providerSeconds into distinct concerns so that long-running, repo-size-dependent work (indexing, DB seed/clone) is never killed by a setup-oriented timeout.

Highlights

• Renamed providerSeconds → providerSetupSeconds (default 1800) for provider image build / dependency install. The old key now raises a clear ConfigError pointing at the new name.
• Docker control ops (start/stop/status/inspect) now use a fixed 120s internal cap so hangs fail fast; this is no longer user-configurable.
• Indexing and DB seed/clone (CGC index --force, CGC bundle export/import, grepai pg_dump/restore) are never capped, because they scale with repo size.

Mechanics

• "Unlimited" is the integer sentinel 0 (round-trips through the --timeout CLI arg); command_runner maps timeout <= 0 to subprocess timeout=None at the call boundary.
• timeoutCaps now accepts 0 (= unlimited) and rejects only negatives.
• Updated example settings and reference docs.