Various perf fixes to minimize linter CPU impact

[anthony-c-martin]

Closes #3785

Changes:

• Allow semantic models to be reused between compilations if the source file has not been modified and contains no dependencies on a source file which has been modified. This avoids constant re-running of linters (and other semantic analysis) across multiple module dependencies on every keypress.
• De-duplicate semantic models between active compilation contexts if multiple files part of the same compilation are open. This minimizes the number of times the linters (and other semantic analysis) must be run.
• Avoid processing resource type strings & module paths in the no-hardcoded-env-urls linter.
• Add debug tracing framework for language server and CLI.

[anthony-c-martin]

@miqm - I'd be very interested to see what sort of difference this change makes to performance with your file structure - if you get a chance to try it out.

Here's a link to the VSIX file from this PR: https://github.com/Azure/bicep/suites/3389732759/artifacts/79971841

The main changes I made:

• Avoid regenerating semantic models for files which have not been changed, or contain no (direct or indirect) references to files which have changed.
• De-duplicate semantic models across different compilations - e.g. if I've got a parent module and child module both open in the editor, there should be only one copy of the child semantic model.

I also added some LSP tracing so that you can see which semantic models are reloaded in real time.

[miqm]

I’ll check it on Monday.

[majastrz]

singleLine

Should exclude the "TRACE: " prefix from what gets logged in msbuild.

[majastrz]

AWESOME!

[miqm]

@anthony-c-martin since it's merged I've downloaded latest main branch vsix and performance is awesome. looking forward for the release.