[libc++] Further refactor sequence container benchmarks (#126129)

This patch does not significantly change how the sequence container
benchmarks are done, but it adopts the same style as the associative
container benchmarks.

This commit does adjust how we were benchmarking push_back, where we
never really measured the overhead of the slow path of push_back (when
we need to reallocate).

Author: ldionne