perf: Performance improvements

[TimBeyer]

What this PR does / why we need it:

This PR adds a bunch of performance improvements, and replaces Bluebird with native promises.

Bluebird

Let's start off with Bluebird: The bluebird repo itself recommends moving to native promises and bluebird is no longer seeing any changes. Meanwhile native promises have seen big performance improvements and also usually show up better in stack traces and in performance profiles.

Repo scan tree indexing

The main performance improvement comes for GARDEN_GIT_SCAN_MODE=repo.
It now indexes the list of files into a tree structure for faster access yielding large improvements in graph resolution speeds for large projects.

Here's a comparison between the current main version and the updated one, run on an M2 mac on a repo of 11032 files with 500 modules.

Using main:

time GARDEN_GIT_SCAN_MODE=repo gdev validate
Validate ✔️

Project is configured with `apiVersion: garden.io/v0`, running with backwards compatibility.
ℹ garden               → Running in Garden environment dev.default
Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
ℹ providers            → Getting status...
✔ providers            → Cached (took 2.1 sec)
ℹ providers            → Run with --force-refresh to force a refresh of provider statuses.
ℹ graph                → Resolving actions and modules...
ℹ graph                → Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
✔ graph                → Done (took 16.5 sec)

OK ✔️
GARDEN_GIT_SCAN_MODE=repo ~/Development/garden/garden/bin/garden validate  76.32s user 7.61s system 133% cpu 1:02.88 total

Using this PR:

time GARDEN_GIT_SCAN_MODE=repo gdev validate
Validate ✔️

Project is configured with `apiVersion: garden.io/v0`, running with backwards compatibility.
ℹ garden               → Running in Garden environment dev.default
Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
ℹ providers            → Getting status...
✔ providers            → Cached (took 2.1 sec)
ℹ providers            → Run with --force-refresh to force a refresh of provider statuses.
ℹ graph                → Resolving actions and modules...
ℹ graph                → Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
✔ graph                → Done (took 8 sec)

OK ✔️
GARDEN_GIT_SCAN_MODE=repo ~/Development/garden/garden/bin/garden validate  64.68s user 6.62s system 136% cpu 52.410 total

As you can see the overall performance improvements are still moderate, but the graph resolution is almost 2x faster.
This makes the repo scan mode the most performant option especially when running in our single binary build which seems to have a lot of overhead for the default scan mode.

For comparison here are both scan modes with our current 0.13.12 binary version on the same repo:

time GARDEN_GIT_SCAN_MODE=repo garden validate
Validate ✔️

Project is configured with `apiVersion: garden.io/v0`, running with backwards compatibility.
ℹ garden               → Running in Garden environment dev.default
Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
ℹ providers            → Getting status...
✔ providers            → Cached (took 2.2 sec)
ℹ providers            → Run with --force-refresh to force a refresh of provider statuses.
ℹ graph                → Resolving actions and modules...
ℹ graph                → Scanning repository at /Users/tim/Development/garden/garden-large-repo-generator
✔ graph                → Done (took 15.9 sec)

OK ✔️
GARDEN_GIT_SCAN_MODE=repo garden validate  78.09s user 8.10s system 138% cpu 1:02.22 total

time garden validate
Validate ✔️

Project is configured with `apiVersion: garden.io/v0`, running with backwards compatibility.
ℹ garden               → Running in Garden environment dev.default
ℹ providers            → Getting status...
✔ providers            → Cached (took 2.1 sec)
ℹ providers            → Run with --force-refresh to force a refresh of provider statuses.
ℹ graph                → Resolving actions and modules...
✔ graph                → Done (took 26.2 sec)

OK ✔️
garden validate  88.09s user 28.67s system 160% cpu 1:12.63 total

You can see that the repo scan mode shows identical performance to the version running just on node for development.
Meanwhile graph resolution on the default scan mode with the binary takes 26 seconds to resolve the graph while just on node it's around 15 seconds (not in the outputs above since I didn't want to add even more noisy text here).

So the new implementation with repo mode should outperform the default scan mode in the binary by a good 3x still when it comes to graph resolution.

GC Improvements

We increase max-semi-space to 64M which should lead to less GC pauses and better performance at the cost of slightly higher memory consumption.

See https://github.com/nodejs/node/blob/main/doc/api/cli.md#useful-v8-options
Also see https://www.alibabacloud.com/blog/better-node-application-performance-through-gc-optimization_595119 and nodejs/node#42511 for some details on impact.

Which issue(s) this PR fixes:

Fixes #

Special notes for your reviewer:

[TimBeyer]

Here I wasn't sure if we normalize everything to POSIX paths also for glob compatibility, or if we do need to use path.sep so that windows paths get split correctly as well.

That probably would also impact the .startsWith part above.

[vvagaytsev]

Wow! This is super awesome! 🚀 Huge thanks for such a great job! ✨
I need more time to take a closer look in review it in detail :)

[vvagaytsev]

Amazing work! Thank you! I've left a few non-blocking comments, LGTM! 💯

[vvagaytsev]

💯 🚀