Inconsistent behavior across using `map`, `broadcast` and `for/loops` and `FixedSizeArrays`

[el-oso]

While investigating a performance difference between Vector and FixedSizeVector, I discovered that there was also a performance difference between map, broadcast and for/loops.

I am not 100% sure that all implementations are equivalent, but I made my best effort.

From the table below you can see that for simple calls stock Vectors perform equally except if the is a do block, but are a mixed bag for FixedSizeVectors
For more complicated things it is a mixed bag and for the full case that I am interested clearly a for/loop behaves differently between Vector and FixedSizeVector. At least on my computer.

With Vector

ID	time	GC time	memory	allocations
["Full", "Generated", "Vector"]	971.000 ns (5%)
["Non-Simple", "Broadcast", "Vector"]	5.570 μs (5%)
["Non-Simple", "Loop", "Vector"]	991.000 ns (5%)
["Non-Simple", "Map", "Vector"]	5.621 μs (5%)
["Simple", "Broadcast", "Vector"]	641.000 ns (5%)
["Simple", "Loop", "Vector"]	641.000 ns (5%)
["Simple", "Map Do", "Vector"]	3.937 μs (5%)
["Simple", "Map", "Vector"]	641.000 ns (5%)

With FixedSizeVector

ID	time	GC time	memory	allocations
["Full", "Generated", "FixedSizeVector"]	5.370 μs (5%)
["Non-Simple", "Broadcast", "FixedSizeVector"]	6.051 μs (5%)
["Non-Simple", "Loop", "FixedSizeVector"]	5.019 μs (5%)
["Non-Simple", "Map", "FixedSizeVector"]	6.001 μs (5%)
["Simple", "Broadcast", "FixedSizeVector"]	641.000 ns (5%)
["Simple", "Loop", "FixedSizeVector"]	3.536 μs (5%)
["Simple", "Map Do", "FixedSizeVector"]	3.917 μs (5%)
["Simple", "Map", "FixedSizeVector"]	641.000 ns (5%)

I include an MWE to replicate these results.

MWESLEEFPiratesNFSA.tar.gz