This is a hybrid between node-haste1 and node-haste2 and currently only works with both of them hooked up. This is terrible.

* _nodeModuleNameToPath: _resolveNodeModuleName
* _moduleNameToPath: _resolveModuleName
* Remove package.json resolution, resolve.sync
  does this.

…ildLoader is called recursively).

* Extract dependencies from `requireActual` and `requireMock`

(this adds two TODO items to make these two methods work properly again, but I’m unblocking myself by stubbing them out for now).

…aphs instances.

* Coverage is now collected for all modules except for node_modules; this simplifies the runner/loader quite a bit.
* Cleaned up a few tiny bits here and there.

…onger. Remove object-assign.

* Look up changed files starting from test files rather than changed files
* Deduplicate testPathDirs and changed test files
* Use the git root dir for tests, not the test path dir so we get correct absolute paths (—onlyChanged *never* worked for jest itself, what!?)
* This might not perform as well as it did before; we’ll see and I might rewrite this again, but for now this works pretty well.
* Simplify the tests. I’m going to make them more robust in a follow-up but most of the things the tests were covering is handled by DependencyGraph

… moduleMap upfront in the TestRunner.

+ minor cleanups to the TestRunner.

Jest now has an actual software architecture. lol.

* use testPathDirs instead of rootDir
* Fix TestRunner-test

* Resolving the entire Graph for every single test file has proven to be unfeasible in a sufficiently large codebase. This change now circumvents all static resolution and falls back to the same dynamic resolution that node-haste1 was providing in Jest. Jest pulls out the entire HasteMap of node-haste2 and uses that. node-modules are resolved inside of Jest only.
* Instead of providing local moduleMaps per test, the entire haste map is now written to disk and read by workers and shared with all tests.
* Jest is now by default using graceful-fs and provides it to test files as well. This is safer to not run into the file limits

The node-haste2 way of resolving shallow dependencies is way too slow for large codebases. The haste reverse-dep-graph lookup is less correct (ie, doesn’t work well with relative modules) but it works well with a haste only system.

Reviewed By: voideanvalue

Differential Revision: D2629814