Leave a cell the full disk stopped out of the matrix and the reports

[tamnd]

A cell that failed because the build machine ran out of disk says nothing about the kernel or the compiler. search already treats it as not run, but the matrix and the release reports still showed it as a failure. In reports/new-gcc-16.2.md, for example, 7.2.9 on i386 reads as GCC 16.2 failing at L2, with error writing to /tmp/ccc2h2gi.s: No space left on device as its first error, and 6.12.112 and 7.2.9 on arm64 fail the same way.

no_result in publish.rs names that case, a failed cell for which build::disk_full holds. matrix() and report::cells() now leave such cells out. An older cell of the same crossing takes the square if there is one, and otherwise the square reads not run until a search runs the cell again. Cells that went over their budget are left as they were, since over-budget classifies them and they still count as edges.

New test: a_cell_the_full_disk_stopped_is_no_result. fmt and clippy are clean on server3 and all 130 tests pass.