smallvec performance issues on common store operations

[teh-cmc]

See #1747 for detailed benchmarks.

The obvious solution seems to be to simply switch to tinyvec which:

• doesn't exhibit the same issues
• doesn't have any unsafe code
• has more features

The only downside is the Default requirement: for those cases where Default is not an option, a vanilla Vec seems like the best bet.

[Wumpf]

a vanilla Vec seems like the best bet.

not necessarily though I'd argue: Sure your benchmark shows quite a few issues on access performance, but I'd expect SmallVecs that stay in their "small bound" to still have way better behavior access behavior on cold cache (which is always a great claim to put out there because it's so hard to prove, heh) than Vec. I mean, all the reasons for why one would like to avoid the heap allocation/Vec are still there obviously

[teh-cmc]

This is appropriately documented and benchmarked now so this issue effectively fixed; each system will have to act accordingly depending on their tradeoffs. Closing.