fix: 🐛 use proper webpack hook

[austinw-fineart]

Refactors the webpack plugin to use the proper hook, only extracting keys from processed files. Also avoids triggering compilation twice in watch mode.

PR Checklist

Please check if your PR fulfills the following requirements:

• The commit message follows our guidelines: https://github.com/jsverse/transloco/blob/master/CONTRIBUTING.md#commit
• Tests for the changes have been added (for bug fixes / features)
• Docs have been added / updated (for bug fixes / features)

PR Type

What kind of change does this PR introduce?

[x] Bugfix
[ ] Feature
[ ] Code style update (formatting, local variables)
[x] Refactoring (no functional changes, no api changes)
[ ] Build related changes
[ ] CI related changes
[ ] Documentation content changes
[ ] Other... Please describe:

What is the current behavior?

Issue Number: N/A

What is the new behavior?

Does this PR introduce a breaking change?

[ ] Yes
[x] No

Other information

[shaharkazaz]

@austinw-fineart Tnx for the PR, I'll take a look soon!

[austinw-fineart]

Made an error with the webpack types so I ended up reusing the stringified contents for RawSource, which means I ended up not using the file utils.

[austinw-fineart]

Ideally being able to reference the webpack types would be preferrable but that'd take a dev or peer dependency so it's understandable.

[shaharkazaz]

@austinw-fineart I'm unfamiliar with testing such plugins, but I want to ensure that regressions are introduced. How did you test it?

[austinw-fineart]

I used my own project to test with. I could add a stripped down version here but it wouldn't exactly match your existing tests. It'll require a dependency on @angular-builders/custom-webpack to run however.

[austinw-fineart]

There is a breaking change if you consider that before the plugin would run extract on the entire project initially, but doing so triggers a second compilation in watch mode. Now the plugin only extracts keys from modified files which makes it more suited as a watch companion rather than a build step which the CLI can fulfill. It's still possible to scan the entire project manually on the first pass but I'm not sure if that's worth the tradeoff.

[shaharkazaz]

@austinw-fineart What's the tradeoff here? Did you feel any big impact when using the plugin as is? The extraction should be pretty quick. Can you share the considerations and metrics (if some were collected)?

[austinw-fineart]

I wouldn't know how to do any synthetic benchmarks but here are my numbers from a large size project:
ℹ 1110 keys were found in 1142 files.
Initial total: 12.56 MB

Version	Run1	Run1²	Run2	Run2²	Run3	Run3²	Mean	Error	StdDev
old	247876 ms	99288 ms	237235 ms	91887 ms	222362 ms	93602 ms	235824.333 ms	7398.9534 ms	12815.3632 ms
new	233102 ms	-	224513 ms	-	218590 ms	-	225401.667 ms	4212.7518 ms	7296.7001 ms

² second compilation triggered immediately after by the file watcher