Enhance required phrase rule generation and tagging from license fields

[AyanSinhaMahapatra]

1. could be useful to generate required phrase rules from the license.homepage_url (but we need to remove a few cases like URLs to files in github/vcs, and lower the rule relevance for these rules)
2. from license.name : We already do this, but it might be more useful to also add required phrase rules removing the versions at the end of names

Note: these will improve coverage for rules in license tags/references and of course these rules will have reduced relevance because URLs can change and licenses can have new versions.

[Kaushik-Kumar-CEG]

hi @AyanSinhaMahapatra
i was looking at how this fits into the current pipeline and had a few quick thoughts:

stripped names like "GPL" would map to multiple license expressions (like gpl-2.0 and gpl-3.0). rather than building a new flow, could we just extend generate_required_phrase_variations() to handle this? it currently just drops the word "the", so it seems like a clean place to put it and handle those collisions

for URLs and for those github/vcs links we want to avoid, maybe we could just autoset the skip_for_required_phrase_generation flag on them? that way they don't pollute the database at all, rather than generating a rule and lowering its relevance

also the reduced relevance part connects to philippe's issue #3945 . if these new rules have lowered relevance, does that implicitly mean they should act as ranking signals rather than hard filters?

would love to hear your thoughts :)

[Kaushik-Kumar-CEG]

I hit this exact issue (URLs) while running some ml tests for the project proposal

in a 300-rule test dataset, url-only rules caused a 57% false positive rate (the model just tags the url itself). I'd say auto-setting that skip flag is a good way to cut the noise right at the source