Stop repeating the index check and the history summaries

Stop repeating the index check and the history summaries

The request memo covered the repository, the open counts and the change
list, but two more reads were repeated:

- The index refresh (issue rows, filters, `#42` numbers) ran on every
  surface that shows one, and each run started by resolving the metadata
  ref. A page calls it from the header, the number lookup and its own
  view, so it now happens once per request, and the resolved commit is
  shared with the change-request read instead of being resolved twice.
- The overview walked the repository's whole history on every load:
  `shortlog` for contributors, `rev-list --count` for the code page's
  commit count, and the tree grouping for languages. Those are properties
  of a commit, and a commit id is content-addressed, so they cannot go
  stale: they are remembered by commit id in a bounded per-process cache
  that evicts oldest-first.

On the fixture (30 change requests, 30 issues) the overview drops from 19
git calls to 9 on a repeated load, the code page from 12 to 11, and every
page loses one `rev-parse` for the metadata ref.