docs(m177): the gate's write half lands, and Write 5 is decided on evidence

The Write 5 entry claimed documentation drift. A repo-wide trace shows the
opposite: `execution_id` has zero production writers of a value, zero
production readers, and one statement that only ever sets it NULL — so the
diagram describes a feature never wired, not a diagram over-describing the
code. Recorded with the evidence, the decision not to port it, and the reason
that actually holds: it would duplicate `fleet.runner_leases` without the
fence or the TTL that make that row trustworthy.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>