The previous `memchr` loop had the fatal flaw that it would break out
of the SIMD routines every time it hit a newline. This resulted in a
throughput drop down to ~250MB/s on my system in the worst case.
By writing SIMD routines specific to newline seeking, we can bump
that up by >500x. Navigating through a 1GB of text now takes ~16ms
independent of the contents.
Since `offset_from_unsigned` has been stabilized,
we can remove simd::distance now.
---------
Signed-off-by: Thallium Megalovania <ferris@thallimega.win>
Co-authored-by: Leonard Hecker <leonard@hecker.io>
