Entropy leak: not removing semicolons about ktfmt

[cushon]

There are some built-in assertions that formatting only changes whitespace, and doesn't add or remove tokens, because it helps guarantee that reformatting will never affect semantics.

There are some similar things we'd like to do for Java, but so far I haven't had great ideas for supporting non-whitespace fixes without adding complexity or regressing guarantees around not making semantic changes.

Another option would be do to this is a separate pass, similar to how import fixes are handled.

[strulovich]

I'm leaning towards adding an extra pass that would be more aggressive and modify the tree. This will also let us supply a flag for it. I'm against configurations, but removing symbols might be too aggressive for people that call this when they save a file, so it might be useful to support.

@cgrushko, @cushon, how does that sound? We could experiment with ; removal first.

[cgrushko]

I was surprised to find out that it wasn't simple to do :)
The underlying engine (google-java-format's) makes some assumptions about tokens not changing. @cushon might be able to provide more details (I forgot the explanation he previously gave me :|)

(Maybe if we modified the tokenizer to not emit ; tokens at all? might be an interesting idea.)

[cgrushko]

Sounds good.
I think we can go ahead with removing ; without a flag.
If we have luck, kotlin-compiler will just tell us that a symbol is redundant.