Use mmap to read filters

[ctz]

This is probably not a serious proposition (converting valid filesystem alterations into memory safety issues isn't great) but acts as a measurement that upki is performance-sensitive in this path.

I have some ideas about how to improve this in a more robust way that need a bit of investigation/measurement.

revocation-check        time:   [2.9656 ms 2.9855 ms 3.0063 ms]
                        change: [−51.569% −50.952% −50.331%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  3 (3.00%) high mild

[codspeed-hq]

Merging this PR will not alter performance

✅ 4 untouched benchmarks

---

_{Comparing jbp-mmap (6201ad4) with main (d18ff83)}

[djc]

Seems bad that CodSpeed doesn't detect an improvement here?

[ctz]

Yeah unfortunately it doesn't measure time in syscalls. Though I'd really expect the large allocation for the fs::read buffer to feature somewhere, but I think it is being drowned by the deserialisation allocations (a good learning for #33 maybe)

[ctz]

I think this proved a point, which is that we were IO-sensitive in revocation checks. See #89 for an alternative approach.