Optimize wasmi::Linker host function setup

[Robbepop]

Closes #979.

cc @athei

Most inefficiencies are in StringInterner::get_or_insert and in memcmp calls to compare definition names.

Findings

• By using an empty "" name for the module name for all host functions we can improve performance by ~15%.
  • This is likely due to a fast check for empty-string to avoid having to use memcmp.
• Introduction of struct LenOrder(Arc<str>) and struct LenOrderStr(str) wrapper types with new Ord implementation to avoid memcmp are very effective, improving performance by roughly 50%. Unfortunately this requires a single line of unsafe Rust code since Rust's BTreeMap does not support custom comparators.

[codecov]

Codecov Report

Attention: Patch coverage is 64.46281% with 43 lines in your changes are missing coverage. Please review.

Project coverage is 81.89%. Comparing base (6c58a0f) to head (92aea8d).

Files	Patch %	Lines
crates/wasmi/src/linker.rs	62.83%	42 Missing ⚠️
crates/wasmi/src/func/mod.rs	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #989      +/-   ##
==========================================
+ Coverage   81.81%   81.89%   +0.08%     
==========================================
  Files         262      262              
  Lines       24110    24186      +76     
==========================================
+ Hits        19726    19808      +82     
+ Misses       4384     4378       -6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Robbepop]

The new experimental LinkerBuilder API benchmarks show that LinkerBuilder::finish is up to 6 times faster than re-constructing the Linker which is a great speed-up for certain use cases.
The downside is that this requires to build up the LinkerBuilder beforehand which is roughly as slow as building up a normal Linker. Thus if a user now builds all Linker using the new LinkerBuilder but only actually builds up a single Linker this is going to be roughly 20% slower.