Add pseudo-xrefs for implicitly imported entities

[jinwoo]

Related to #6 and #97.

#97 adds xrefs for explicit imports but not for implicit ones.

[robinp]

Re "no trivial span to put these":

That's true, but one could pack an existing span (say the module name span), given a good UI that can handle that.

Re "how is this useful":

One usecase I can think of is showing "imported via" info on the UI. Sometimes, if there are many implicit imports, it is not clear which one imports a given name. If the ref/imports is available on the file, the UI could reconstruct that info.

OTOH if the exported name list is long (maybe auto-generated stuff from protos), this would be quite a burden, and need limiting...

I would say if someone cares (maybe CodeUnderhood one day), they should implement this.

[robinp]

Re the "imported via" feature: I think that is solvable even without having the implicit import refs. One could query the reverse-childof edges of the imported module, and intersect with the refs in the current module. That would automatically yield the actually used imports from a given module.

[robinp]

Inspired by TreeTide/underhood#10, I believe we should consider

• switching the imported names from plain ref/imports to the "aliasing" case described in refactor(typescript_indexer): emit "named" edges for imports kythe/kythe#3934 (comment).

• doing the same for implicitly imported entities (at some arbitrary span?), but only for the entities actually used in the current module (vs all entities exported by the module). This is to control import bloat.

Why do I think this is a better alternative than letting the frontend calculate the imported-via info itself?

• It is more explicit - less logic left for the frontend.

• It is more in line with other indexers - it is expected that Typescript, Python etc indexers would have similar behavior.

• If the frontend wanted to implicitly guess imported-via (based on matching an imported module name with the module name of a used name), it would run into trouble with reexports: if module M reexports foo from module X, but we only import M and use foo, the implicit guess won't work (there's no module X, which is the module of foo, directly imported).

Implementation Note: this can be implemented purely in the kythe-frontend, shouldn't affect GHC traversal.