perf: Use serde to aid in loading docs

[WillLillis]

This PR addresses #75 by changing how data is loaded at server start. Before, each xml file was parsed at server start into a vector of Instructions, Registers, or Directives. Now, the following takes place:

• A separate binary (asm_docs_parsing) parses said xml files into vectors and then serializes them once beforehand. The serialized content is then written to a separate file
• At server start, the serialized file is deserialized, and the rest of the server's operation proceeds as normal

Before marking this PR as ready, I'd like to do a thorough performance measurement, as well as to write some tests to ensure the serialized docs stay in sync with their xml counterparts.

Averaged over 10 trials each, the server now starts about 12% faster, which is an absolute win. The performance gains added here will continue to help as we add additional architectures as well.

However, some initial measurements are promising:

• The x86 register set showed a 1.88x loading time improvement
• The x86 instruction set showed a 1.69x loading time improvement
• Raw and serialized data file sizes are all in the same ballpark, with an aggregate reduction of 736KB. See below for details:

	x86 Instructions	x86_64 Instructions	z80 Instructions	x86 Registers	x86_64 Registers	z80 Registers	Gas Directives
Raw Size (bytes)	4882695	5241289	117347	55149	101735	5047	57010
Serialized Size (bytes)	4557857	4754451	177181	58589	111965	6125	57107
Change	-324838	-486838	+59834	+3440	+10230	+1078	+97

Closes #75

[lu-zero]

If you store it as a bincode, it could lead to even bigger gains.

[WillLillis]

If you store it as a bincode, it could lead to even bigger gains.

Thanks, wasn't aware of that crate!

#93