Much faster push_ascii for NEON, add bench

[imartayan]

No description provided.

[RagnarGrootKoerkamp]

Updated the bench for multiple lengths and to reuse the allocation, and added & fixed a test for BitSeqVec::push_ascii which was broken when more than 1 sequence of non-multiple-of-8 length was pushed.

This is what I now get on x86:

test test::push_ascii_1bit_bench ...
Bench BitSeqVec::from_ascii
Len     100 => 2.441 Gbp/s
Len     150 => 3.043 Gbp/s
Len     200 => 3.368 Gbp/s
Len    1000 => 4.626 Gbp/s
Len 1000000 => 5.117 Gbp/s

test test::push_ascii_2bit_bench ...
Bench PackedSeqVec::from_ascii
Len     100 => 3.074 Gbp/s
Len     150 => 3.426 Gbp/s
Len     200 => 4.466 Gbp/s
Len    1000 => 5.373 Gbp/s
Len 1000000 => 5.653 Gbp/s

After some more commits to reduce bounds checking:

test test::push_ascii_1bit_bench ... 
Bench BitSeqVec::from_ascii
Len     100 => 3.672 Gbp/s
Len     150 => 5.311 Gbp/s
Len     200 => 6.510 Gbp/s
Len    1000 => 15.963 Gbp/s
Len 1000000 => 16.289 Gbp/s

test test::push_ascii_2bit_bench ... 
Bench PackedSeqVec::from_ascii
Len     100 => 4.616 Gbp/s
Len     150 => 5.472 Gbp/s
Len     200 => 8.919 Gbp/s
Len    1000 => 16.533 Gbp/s
Len 1000000 => 15.528 Gbp/s