Delete board_from_cache's dead state field (#597)

[fdaviddpt]

What

board_from_cache in scripts/statusline.py computed a state field (unknown/partial/measured) that nothing downstream ever read -- _board_field, the renderer, reads prs/issues/issues_external/checks directly and renders ? per missing value regardless of what state said. #595 had just added a rule to state (a cache with issues but no issues_external is partial) that, for the same reason, could never affect anything a maintainer sees.

#597 named two directions and deliberately chose neither: give state a reader (the render), or delete it. This takes the delete path.

Why delete rather than wire a reader

The render is the most-visible field on the line and a design change there (a partial board rendering differently from a measured one) needs its own argument about what it costs in width and reader attention -- and, on inspection, state was never even a complete summary of what the render already shows: it only ever considered prs/issues/issues_external, never checks. A reader added for it would have inherited that gap. Each of the three counts state summarized already carries its own missing/present distinction (None vs an int), which is exactly what _board_field already renders as ? vs a number. A summary that can disagree with the values it summarizes -- and, in this case, was already known to be inert -- is a second copy of the same fact rather than new information.

Verified the premise

Grepped board_from_cache and ["state"]/.get("state") across scripts/, commands/, hooks/: the only production caller of board_from_cache is gather() (scripts/statusline.py:1479), which never reads board["state"]; the board dict only reaches _board_field, which reads the four other keys via .get(). No other caller anywhere in the repo reads this field. Confirmed by both spawned reviews independently.

Tests

Red: tests/test_statusline_state_deleted_597.py::test_board_from_cache_carries_no_state_key failed before the fix (state was present). Green after. Positive control in the same file (test_the_per_field_questionmarks_still_distinguish_every_case_state_used_to_name) proves the per-field ?/int distinction alone still recovers every case state used to name, including the must-fire zero-vs-missing control for issues_external. Three existing test files (test_statusline_479.py, test_statusline_board_external_595.py) had their board["state"] assertions removed as a deliberate contract change, not an oversight -- state is deleted, so there is no contract left to assert.

Full suite: 3766 passed, 7 skipped (pre-existing, unrelated), coverage 89.34%.

Docs

README.md does not document the board field format at all (confirmed, matches the issue's own note) -- no change needed. .oss/statusline.py, the owned vendored copy, is rendered from scripts/statusline.py at scaffold time (scaffold.py::_owned_statusline), not hand-maintained -- it will carry this fix at the next re-scaffold, noted in the changelog fragment.

Closes #597