Connect MCP servers concurrently at startup

[Aaronontheweb]

Change

• Start all enabled MCP connections through their existing server gates.
• Wait for every initial attempt before the daemon listener starts.
• Keep each failed server status separate from successful server catalogs.
• Update SPEC-011 and the operations skill.

Evidence

• The new two-server startup test failed against the serial baseline, then passed after this change.
• The daemon test project passed: 1,134 passed, one skipped.
• The Spark2 Skill Discovery eval passed: 8/8 cases, one run each.
• Slopwatch found zero issues. Header verification and diff checks passed.
• Three isolated four-server boots per image used the same config and Spark2 endpoint.
• The span between first and last MCP connections fell from 56-65 ms to about 1 ms.
• The median span from memory warmup to the last MCP connection fell from 227 ms to 169 ms.

The fixtures connect quickly, and total boot time varied between runs. These traces do not establish the gain on the active daemon. I did not restart that daemon.

Stack

This PR can merge independently of #2202. Both PRs update the operations skill. Resolve the version line as 2.74.9 after #2202 merges.

[Aaronontheweb]

LGTM

[Aaronontheweb]

Goal of this PR is to improve Netclaw's startup time following a restart - along with the work we're doing on making shutdowns happen faster too