perf(git): use sub-tree scan mode for config files scan

[vvagaytsev]

What this PR does / why we need it:

This idea was discussed in #6461.

Use sub-tree scan mode for config files scanning. Now Garden has the second VCS handler - a dedicated SubTreeGitHandler that only performs config files scan. Both VCS handlers share the same TreeCache.

Which issue(s) this PR fixes:

Fixes #

Special notes for your reviewer:

[stefreak]

The big question for me here is: what are the benchmark results? is it faster, slower, same performance?

[edvald]

To evaluate that we need to look at various cases, even different commands etc. I can't quite imagine how this would turn out to be slower outside of very rare cases and even then marginally, but I don't object to running a few performance tests to see how it shakes out.

[vvagaytsev]

@stefreak sure, sorry, I accidentally created it before I completed the description.

With sub-tree scan mode it doesn't matter if we enable or disable the untracked files hashing implemented in #6461, it's equally fast in either way. The Garden repo scan example from #6461 (with node_modules included) gets scanned in about 0.6 sec.

With repo scan mode and disabled hashing for untracked files, the results are almost the same: about ~0.6 - 0.9 sec.

I think that's because repo does process more files than sub-tree. Processing without hash calculation is very fast, but processing less files is even faster.

We can keep both changes:

1. Keep a dedicated VcsHandler for config scan and use sub-tree - very fast.
2. Keep an option to disable hashing of untracked files - then we can also use repo mode for config scan if we have to do that for some reason.

[vvagaytsev]

@edvald @stefreak thanks for the comments!

Sure, let's do more performance tests and benchmarks with different types of projects. I've converted this to draft.

[edvald]

No need to overdo it @vvagaytsev :) Just good to see a couple of different variations.

[stefreak]

I agree, if y'all think this is definitely an improvement I'm all for it, I just don't understand the vcs code to be honest so that's why I wanted to see proof that this is better than before :)

[vvagaytsev]

Just tested this against main branch with 100k files project (used https://github.com/garden-io/garden-large-repo-generator to generate that). All files were tracked.

This PR did the initial scan in 0.2-0.3 sec against 0.5-0.6 sec in main. So, it looks it's ready to go.