refactor(linter): severity and recommendation of lint rules

[ematipico]

Summary

• make the existing recommended set of rules less pedantic (the changeset presents a summary of it);
• noForEach isn't recommended anymore because performance is debatable (it depends by the runtime) and it changes with time;
• all the *useless* rules emit an information diagnostic, because they trigger ... unless code! no need to block the CI for it;
• noDelete isn't recommended anymore because it's the way to for environment variables in Node.js, so Biome should be less pedantic about it;

Test Plan

Many snapshot tests are updated because our test infra was hardcoding the severity of our rules. Now that rules have their own severity, there's no need for that anymore.

[codspeed-hq]

CodSpeed Performance Report

Merging #5133 will not alter performance

_{Comparing refactor/default-severity (33ff868) with main (d739883)}

Summary

✅ 94 untouched benchmarks

[arendjr]

Nice!

[Conaclos]

Why this rule is no longer recommended?

[ematipico]

clippy has this rule under the pedantic category, and it doesn't recommended it, I think we should follow their lead and stick with it.

Our recommendation set is very opinionated and we should reduce it. What do you think?

[Conaclos]

Many users like the fact that Biome is opinionated.
I accept that we could be more cautious about what we add in the recommended rule set.
Personally I could keep this rule in the recommended set because it brings one of the mission of Biome: making dev better dev by helping to learn JS API.
We could reduce the severity level to information.

[ematipico]

Sounds good!