add isbitstype optimized path for FixedSizeList getindex

[jrevels]

ref #103 (comment)

#103 before this PR:

julia> @time copy(uuids_fixed);
  0.142450 seconds (1.06 M allocations: 74.713 MiB, 2.43% gc time)

#103 after this PR:

julia> @time copy(y.recording_uuid);
  0.016563 seconds (2 allocations: 3.248 MiB)

Order of magnitude improvement + allocations/memory now on par with pre-#103 behavior. Nice!

Still an order of magnitude of runtime perf on the table compared to the equivalent Primitive though. will play around and check whether branching changes/direct pointer manipulation might recover some of that.

[codecov]

Codecov Report

Merging #104 (2d8b8d8) into main (31476c9) will decrease coverage by 0.42%.
The diff coverage is 96.15%.

@@            Coverage Diff             @@
##             main     #104      +/-   ##
==========================================
- Coverage   84.48%   84.06%   -0.43%     
==========================================
  Files          22       23       +1     
  Lines        2611     2692      +81     
==========================================
+ Hits         2206     2263      +57     
- Misses        405      429      +24     

Impacted Files	Coverage Δ
src/arraytypes/fixedsizelist.jl	85.71% <93.33%> (-1.63%)	⬇️
src/arrowtypes.jl	90.00% <100.00%> (+0.25%)	⬆️
src/arraytypes/list.jl	88.59% <0.00%> (-1.59%)	⬇️
src/arraytypes/map.jl	95.38% <0.00%> (-1.50%)	⬇️
src/arraytypes/bool.jl	85.24% <0.00%> (-1.43%)	⬇️
src/arraytypes/unions.jl	82.06% <0.00%> (-0.57%)	⬇️
src/metadata/Flatbuf.jl	100.00% <0.00%> (ø)
src/utils.jl	82.79% <0.00%> (+1.27%)	⬆️
src/Arrow.jl	52.38% <0.00%> (+5.01%)	⬆️
... and 1 more

---

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 31476c9...2d8b8d8. Read the comment docs.

[jrevels]

Using the same _unsafe_cast strategy from #103, we now get:

julia> @time copy(y.recording_uuid);
  0.009492 seconds (2 allocations: 3.248 MiB) 

Still slower than the Primitive direct copy but now within the same order of magnitude, so I'm happy 😁

Once either #103 or this PR is merged, the remaining PR should be refactored to use the same underlying _unsafe_cast! utility defined here. (shouldn't matter which one is merged first)

[quinnj]

LGTM, let's just rebase the _unsafe_cast out since #103 is merged

[jrevels]

let's just rebase the _unsafe_cast out since #103 is merged

done!