alloc-tls: Consider removing alloc_tls_fast_with! macro

[joshlf]

Microbenchmarks on my laptop now show that TLSSlot::with is just as fast (2 ns/op) as the alloc_tls_fast_with! macro. If memory serves, these microbenchmarks used to show (on this same model of laptop) that TLSSlot::with took 4 ns/op. Thus, it's worth looking into whether elfmalloc's macrobenchmarks no longer show a difference in speed between TLSSlot::with and alloc_tls_fast_with!, and if so, to consider removing the latter.