Use different caching for MIRI runs

[alamb]

Which issue does this PR close?

Closes #879

Rationale for this change

MIRI is failing in some "can not reproduce" way on github.

We have previously seen github fail due to ancient stuff in the cache (e.g. #878), though in that case it was related to the cargo cache

What changes are included in this PR?

Change the cache setting for the MIRI job to:

1. Be different (and thus not use whatever is wrong with the current one)
2. More closely follow the other jobs caching

Are there any user-facing changes?

No

[codecov-commenter]

Codecov Report

Merging #892 (82f2236) into master (898924f) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #892   +/-   ##
=======================================
  Coverage   82.45%   82.45%           
=======================================
  Files         168      168           
  Lines       48231    48231           
=======================================
  Hits        39767    39767           
  Misses       8464     8464           

Impacted Files	Coverage Δ
parquet/src/arrow/array_reader.rs	77.72% <0.00%> (-0.09%)	⬇️
parquet/src/encodings/encoding.rs	94.67% <0.00%> (+0.19%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 898924f...82f2236. Read the comment docs.

[jimexist]

👍

[alamb]

I plan to merge this in once CI passes.

[jimexist]

Alas one of the two true hard problems in computer science.