mech_stats.py overstates what the merged-PR count measures

[realmarcin]

Found reviewing #63, in a docstring that PR introduces.

scripts/fleet/mech_stats.py says the merged pull-request count is "the closest thing the fleet has to a measure of how much human review the corpus has passed through". That claims more than the number supports. The count is every merged PR in a repository's life, and the Mech repositories merge a lot of automation: seeding runs, regeneration, dependency updates and agent-authored changes all land as merged PRs alongside human curation.

The number is still worth showing. It is a real measure of development activity, and the page presents it as "merged PRs" without interpretation, which is fine. It is the docstring that overreaches, and docstrings are where the next person goes to learn what a number means.

Suggested fix: say what it counts, which is every merged pull request in the repository's history, and drop the claim about human review. If a human-review measure is actually wanted, it would need PR authorship or review filtering, which is a separate piece of work.

[realmarcin]

Done on main. The docstring now says the count is every merged pull request in the repository's history, curation and automation alike, and that it measures development activity rather than how much of it was human. Landed in #77.