[core] Refactor CPD integration

[adangel]

Notes originally from #3809:

• reuse common infrastructure like text documents
• CPD programmatic API - goal is to reuse the same abstractions as PMD like FileCollector. No compatibility story planned for master. Also see [core] Provide a CpdAnalysis class as a programmatic entry point into CPD #4204
• Related to both items, maybe the cleanest approach would be to make CPD a PMD rule, and we could remove the programmatic API of CPD (keeping the CLI for convenience)

[jsotuyod]

I don't think implementing CPD as a rule is viable in the short term (if at all).

• it requires statefullness (clashing with [core] Rule API refactoring #4321) or multifile analysis.
• it requires the rule to either be able to cache arbitrary data or automatically invalidates the cache, requiring all files to be analyzed at least for this rule.

[oowekyala]

While it cannot be implemented as a rule yet, the hurdles you mention are the topic of #3920.

[oowekyala]

Also I want to eventually implement tree-based copy paste detection which would work by comparing trees instead of tokens. The most reasonable way to implement this is with #3920, because it still requires parsing files. I have an old prototype which gives much higher-quality results than CPD (https://github.com/oowekyala/pmd/tree/tree-clone-detection). This has the same challenges you mention about incremental analysis, but allowing one rule to process all files is not IMO a problem, as long as those rules that don't need to rerun use the violation cache.