[Record and Playback] Only record the tests that have changed

[sadasant]

To reduce the amount of changes per PR, as well as to keep older untraceable tests clean of the folder tree, we should:

1. Find the set of tests that changed (see note below).
2. Only record the tests that changed.
3. Find the recording files that have no matching test.
4. Remove the files that didn't match any existing test.

Note: To find the changing tests we can either:

a. Use git (either a Node client or child_process) to get the changed files through a diff.
b. Run the playback of each test to see if it doesn't pass, and then add them to the changed-files-pool.
- If so, we need to suppress any console message that might be related to the playback tests before attempting to record them.
- We must also make sure that these playbacks don't add substantial delay to the recordings.

[HarshaNalluru]

Some thoughts on how to implement this -
We can leverage the mocha's this context object to do wonders.
We can customize the console logs in the afterEach sections and in case a test fails in during playback, we generate the recording for the test.

Discussion in mochajs/mocha#1998 might be helpful.

[sadasant]

This might also be sufficient to close this other issue: #4879

[HarshaNalluru]

My proposal is to run the playback mode for the tests and if a test fails, then record it.

[sadasant]

What I want to do (though this will change as I dive into the code):

• Separate the Node Recorder and the Browser Recorder into two files.
• Separate the code that deals with reading and writing files into re-usable logic, if possible.
• Separate the code that deals with running each test into re-usable logic, if possible.
• Write unit tests for the separated methods.
• See how to best fit a mechanism to prevent recording from happening if no changes have happened.
• Write the mechanism that prevents recording from happening if no changes have happened.
  • I'm currently thinking on comparing the test that we're running against a previously recorded hash.
  • If no hash exist, record and generate a hash.
• Write unit tests for the logic that prevents recordings from happening.
• Make sure everything works in all of the dependencies, but don't commit those recordings and hashes to the PR with the changes.
• Make a PR with the new recordings and hashes ONLY.

[sadasant]

Ignore the previous points, here's my draft PR: #7213

My current plan is to finish the unit tests and turn it ready for review.