Make the npm steps roundtrip serializable

[nedtwigg]

• These aren't done yet, but it's close. The gist is that I think these are likely going to implement FormatterStep themselves, no Formatter.createLazy. I think this should be the last PR needed for Support Gradle configuration cache on multiple JVMs  #1274. Too sleepy to finish right now, I'll get back to it...

[nedtwigg]

@simschla Just FYI, to support Gradle's configuration cache, we now have a separate FileSignature and FileSignature.Promised. The deal is that .Promised can roundtrip through serialization losslessly, and then FileSignature has the path-independent equality semantics but it can't roundtrip losslessly.

When you create a FormatterStep, you can now choose between implementing FormatterStep from scratch, or you can provide two separate states - one that gets roundtripped through serialization for configuration cache, and another which is used for equality checks.

In this PR I hacked up the NPM steps by removing all the transient fields so that the steps all work again. The only asterisk is that if you change the contents of a configuration file, these steps will not mark themselves as changed at the moment. I'm tracking that issue in #2140.

[simschla]

@simschla Just FYI, to support Gradle's configuration cache, we now have a separate FileSignature and FileSignature.Promised. The deal is that .Promised can roundtrip through serialization losslessly, and then FileSignature has the path-independent equality semantics but it can't roundtrip losslessly.

When you create a FormatterStep, you can now choose between implementing FormatterStep from scratch, or you can provide two separate states - one that gets roundtripped through serialization for configuration cache, and another which is used for equality checks.

In this PR I hacked up the NPM steps by removing all the transient fields so that the steps all work again. The only asterisk is that if you change the contents of a configuration file, these steps will not mark themselves as changed at the moment. I'm tracking that issue in #2140.

Thank you for the heads-up. I'll look into #2140 as soon as time permits!