Use @definitelytyped/types-registry for ATA

[sandersn]

Use @definitelytyped/types-registry for ATA and fall back to types-registry if npm install @definitelytyped/types-registry doesn't work.

Notes:

1. This currently requires you to be authenticated with npm. After the Github package registry leaves beta, that will no longer be true. This PR should not be merged until GHPR is generally available, but it's useful for testing until that date.
2. The existing error handling is incorrect -- it handles an exception that's already handled, while ignoring the return code of execSyncAndLog. I corrected it so that it logs if either npm or GHPR fails.
3. I haven't added tests for this change yet. I see some references to types-registry in the tests so I'll see if that's the right place.

Step 2 of #33330

Edit: Ignoring whitespace makes the diff a little easier to read.
Further edit: I had to create a new file, so the diff is basically impossible to read. You're better off looking at just the first commit to figure out what the PR does, then reading the rest separately.

[sandersn]

Adding some reviewers for visibility mainly.

The tests all mock the registry itself and override the constructor. (This explains why TypingsInstaller is an apparently superfluous base-class -- it's all the pure code in one place, with NodeTypingsInstaller doing all the mutation.)

Any idea how this could be tested?

[andrewbranch]

For testability, you could move the execSync implementation onto this.installTypingsHost?

[sandersn]

So. The second commit is now unit tested. This was the result of lots of flailing about, so let me explain what happened so nobody has to repeat it.

1. NodeTypingsInstaller is now in typingsInstallerCore, which is the "library" part of typingsInstaller. No instances are created as a side-effect of including typingsInstallerCore in your compilation. That is not true of typingsInstaller. I don't like the typingsInstaller/typingsInstallerCore alternation. I'm considering [-App/-Core] instead, plus a comment at the top of -App explaining that it is intentionally not unit-testable.
2. NodeTypingsInstaller now uses undefined instead of "UPDATE FAILED" as an error marker.
3. execSyncAndLog now returns true for success instead of true for failure. This makes the whole calling tree simpler to read.

[andrewbranch]

Meant to come back around and approve this after the testing changes 👍

[sandersn]

Since this is reviewed, I'm going to merge as-is when GPR comes out of beta, but at our meeting on Friday @minestarks pointed out that we should work with the editor to have it pass a --registry= flag to typingsInstaller so that in case of emergency it can be disabled quickly by shipping an editor update, not a typescript update.

[minestarks]

@sandersn Thanks -- the ability to configure / turn off the registry override is a pretty important requirement from my perspective, can we track it with an issue to make sure it doesn't slip through the cracks by 3.8 beta?

[sandersn]

#35038

[sandersn]

This is blocked until Github Packages allows installation of packages without authenticating with a github user's credentials, which it does not yet. I'll update the bug when that does. In the meantime, I may fold in the fix to #35038 to this PR.

[sandersn]

This may not be needed since presumably npm and github packages will soon be sharing infrastructure and microsoft will be paying the bill for hosting both.

[sandersn]

After discussing this, we decided that this feature doesn't make sense with Microsoft owning both github and npm. There's no gain in redundancy anymore.