flat_array_map appears to be too slow when querying

[tzaeschke]

See #39. the flat_array_map appears to be too slow when querying compared to sparse_map. Confirm and fix.

[tzaeschke]

This may be related to TZCNT/CTZ not being compiled to a CPU instruction, see #103.

[tzaeschke]

-> Probably not CTZ, but somehow lower_bound seems quite expensive, even when done with TZCNT which should be just one cycle with -mbmi2.
is it maybe the waste of memory (due to allocating 2^k entries for usually to entries)? Or the non-linear access(?)?