feat: make AotConverterGenerator incremental and optimise

[TimothyMakkison]

• Add system to make AotConverterGenerator incremental
• Descend root node once instead of descending twice

Before

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
RunGenerator	576.4 ms	10.45 ms	8.72 ms	11000.0000	3000.0000	101.98 MB

After

Method	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
RunGenerator	329.8 ms	5.83 ms	13.04 ms	325.1 ms	11000.0000	3000.0000	101.19 MB

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[TimothyMakkison]

Added AotConverterGeneratorIncrementalTests to verify that incremental caching is occuring.

[TimothyMakkison]

Last update, added assertion that for EquatbableArray<ConversionMetada> length to avoid mistakes I made when debgging these tests.

For the future: might be worth looking at using CreateSyntaxProvider or better yet ForAttributeWithMetadataName instead of descending every syntax tree.
That and using WellKnownTypes, SymbolAccessor (when added) and avoiding calling Members and GetAttributes repeatedly.

[thomhurst]

Yeah definitely. These source generators are definitely not optimized! I was learning them as I went along.

Appreciate the help!