fix: Drop unneeded coverage data from nyc --all

[coreyfarrell]

Frequently users run nyc --all in a way that causes source files to be
transpiled for actual testing but not transpiled for --all. This
produces incompatible coverage data and inconsistantly wrong reporting.

The work around here is to drop coverage produced by --all for any
file where we have coverage produced by actual test runs. This ensures
that we prefer code that was transpiled in the way which tests actually
ran.

Fixes #123, #224, #260, #322, #413

[coreyfarrell]

See istanbuljs/nyc#1155 for nyc side of this.

[bcoe]

This is clever. we should definitely prefer the real coverage over the files we walked for all.

It's weird that we end up with incompatible coverage when using --all, but I can't think of a reason why we ever need to debug this fully?

[coreyfarrell]

I believe it's because we merge coverage data before performing source-map remapping. So --all was producing coverage for the original sources and the test runs were producing coverage data for the transpiled sources. So it's really not unexpected if --all is not running the transpiler.

[JaKXz]

Makes sense to me at a high level... don't know enough of the details to give a more thorough review, sorry!

[nshoes]

@coreyfarrell Do you know when this will be released for istanbul-lib-coverage?

[coreyfarrell]

I don't have an exact date but it probably won't take long for a new alpha release of istanbul-lib-coverage. FYI this update will not apply to nyc 14.x. nyc 15.x is still being worked on, for progress on that you can follow istanbuljs/nyc#1104 (no ETA available).