performance improvements

[hannesm]

to not loose the contents of closed PRs and issues, I collect potential performance improvements here.

@samoht mentioned here:
would be good to make the swap test only once by moving the check at the toplevel, e.g. something like:

  let get_uint64 =
    if Sys.big_endian then get_uint64 "BE"
    else (fun t i -> get_uint64 "BE" t i |> swap64)

Note: I haven't benchmarked this very extensively but it seems to go a bit faster than the current implementation.

@pqwy pasted the assembly of an accessor in #195

@pqwy proposed a benchmark in #194

[hannesm]

a broader, very welcome plan would be to have a performance suite for libraries such as cstruct. is that in scope for the new OCaml build infrastructure @avsm @jpdeplaix