Improve fragment handling

[AMDmi3]

Handle a case where's a lot of links which only differ with fragment part, such as nix package recipe urls for haskell modules which all come from a single file (a lot of urls differing with a line number only):

https://github.com/NixOS/nixpkgs/blob/nixos-unstable/pkgs/development/haskell-modules/hackage-packages.nix#L23075

It doesn't make sense to check them all as they are a single url in fact.

Possible solutions:

• Simple: when checking, group links by fragmentless url. This allows to process a huge batch of new urls quickly, however is not effective solution long-term, as link recheck times are random and not many urls will get into a single batch for recheck.
• Instead of checking, take status from a fragmentless url. However, we may not have one in the database (to have one, it should be mentioned in some package explicitly; it's no the case for example above, for instance). We can add such an url, but to work cleanly it would require more complicated refcounting mechanism, as links would reference other links (however that's useful for redirects too).
• Strip fragments on the package level, e.g. keep fragmentless urls in the links database, and keep fragment part in the package.