allocate small strings in 8-byte-aligned pools

[JeffBezanson]

This is a less-disruptive alternative to #40727 that keeps the representation the same and just uses less alignment padding.

Before:

function example()
    for j = 1:20, i = 1:1_000_000
        string(i)
    end
end

julia> @btime example()
  598.121 ms (40000000 allocations: 1.79 GiB)

julia> @btime for i = 1:2^24
    "a"*"b"
end
  247.768 ms (16777216 allocations: 512.00 MiB)

after:

julia> @btime example()
  607.298 ms (40000000 allocations: 1.64 GiB)

julia> @btime for i = 1:2^24
    "a"*"b"
end
  247.473 ms (16777216 allocations: 384.00 MiB)

@nanosoldier runbenchmarks("string", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @christopher-dG

[JeffBezanson]

We discussed this a bit on triage. Seems like we agree it's a win-win. Plus we could use the new pools in the future for other objects with less-strict alignment, e.g. a struct of 9 bytes.

[oscardssmith]

If we're back-porting the change to array size to 1.7, do we want to back-port this too?