More parallelism + better understandability for snapshot tests

[jaybosamiya-ms]

This PR builds on #95, separating out the various snapshot tests that were previously merged via inline-crate.

[parno]

Looks useful to me. Out of curiosity, how much speed up do you see from the additional parallelism?

[jaybosamiya-ms]

Depends on the number of cores on the machine; quick test on my laptop changed this from ~30s to ~14s.

I was hoping for greater speedup, given that within ~2s, all the non-snapshot tests have completed, but I guess some of the files in the snapshot tests are fairly slow individually so we are likely not going to get much faster:

        PASS [   8.119s] verusfmt::snapshot-examples verus_snapshot_unchanged__01__vstd__arithmetic__div_mod_rs
        PASS [  10.373s] verusfmt::snapshot-examples ironfleet_snapshot_unchanged__07__delegation_map_v_rs
        PASS [  11.876s] verusfmt::snapshot-examples pagetable_unchanged__06__impl_u__l2_impl_rs

A non-trivial reason for testing being slow now is that verusfmt is run in debug mode during testing.

I guess my other main motivation with this change was to have the snapshots not be dependent on inline-crate, and thus actually be more readable (and eventually we migrate to having an easy way for projects to opt into being on our CI with the kind of automated snapshot pulling that we do for Verus's snapshot)