Invalidate instead of flushing the mapped range when reading back BLAS compacted size.

[Vecvec]

Connections
Noticed while looking for the cause of #8825 , but this is not a fix for it.

Description
Previously, if the buffer mapping for a given BLAS's compact readback buffer was non-coherent (I want to say incoherent but I think that's wrong) it would call device.flush_mapped_ranges. flush_mapped_ranges is meant for CPU -> GPU writes, but this is a readback buffer, so it does a GPU -> CPU read and so needs invalidate_mapped_ranges instead.

Testing
I don't think my computers return non-coherent mappings, so probably none.

Squash or Rebase?
Squash

Checklist

• Run cargo fmt.
• Run cargo clippy --tests. If applicable, add:
• Run cargo xtask test to run tests.
• If this contains user-facing changes, add a CHANGELOG.md entry.

[cwfitzgerald]

Good catch!