Allow to compile Wasm modules in host functions called from the Wasmi executor (take 2)

[Robbepop]

Closes #631.

This is another relatively naive fix for #631. It heavily refactors the CodeMap to no longer use atomics for synchronization but a global Mutex. The idea behind this is that this is the lowest we can get with simple synchronization locks. If this technique does not suffice the performance needs, we need to look elsewhere.

Benchmarks comparing this PR with main:

Rust-sourced benchmarks do not significantly regress:

Call-intense benchmark regress as expected:

Other benchmarks are also not signficantly affected:

[codecov]

Codecov Report

Attention: Patch coverage is 41.04046% with 102 lines in your changes missing coverage. Please review.

Project coverage is 80.54%. Comparing base (3016673) to head (9e1b555).

Files	Patch %	Lines
crates/wasmi/src/engine/code_map.rs	31.03%	100 Missing ⚠️
crates/wasmi/src/engine/mod.rs	71.42%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1122      +/-   ##
==========================================
+ Coverage   80.51%   80.54%   +0.03%     
==========================================
  Files         271      272       +1     
  Lines       25110    25088      -22     
==========================================
- Hits        20217    20208       -9     
+ Misses       4893     4880      -13     

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

[Robbepop]

Due to the progress in optimizing the Mutex-based CodeMap implementation provided by this PR and only seeing performance regressions mostly for call-intense workloads (as expected) with mostly single digit percentage numbers, I am confident enough in this implementation to merge it for now with the outlook of finding a better, ideally lock-free and wait-free CodeMap data structure as a follow-up PR to hash out the performance regressions of this PR.