[POC] feat: memchr-n

[hsqStephenZhang]

What's this PR about?

Implement shufti table based approach mentioned in #206 to support memchr-N (N could be arbitrary usize as long as shufti algo allows)

Background

see #206.

Implementation

1. port shuffle_bytes and shift_8bit_lane_right methods from aho_corasick.
2. add memchrn_iter, MemchrN public API

This is still experimental, and it adds many overhead for maintenance. Perhaps a better approach is to create a separate crate named memchr-n to support this feature.

The purpose of this PR(now) is mainly for discussion since its quite a big change. Comments are welcomed. (But don't get trapped in current code style)

[hsqStephenZhang]

if we wrap memchr-n over aho_corasick's Searcher, it works, but has some penalty for unnecessary abstraction.

[hsqStephenZhang]

@BurntSushi Would you like to take a brief look? Do you think this approach is feasible? If so, I could fix the Ci problems and go on work on this PR.