move nuget update analysis to native code

[brettfo]

The accompanying smoke test PR is:

• update smoke tests to reflect nuget updater changes smoke-tests#211

---

This PR adds a new native command analyze that takes the place of the update_checker in Ruby. The purpose of this code is to query the available NuGet package sources to see what updates are available. With this change we now no longer manually poke any NuGet APIs; we're using "real" NuGet bits for all of it which means we shouldn't have any more issues around not-quite-compliant NuGet feeds.

There is also a test change in Ruby to stub the calls into the native C# tool to instead simply dump a pre-baked JSON response. This takes several minutes off of the CI time.

---

Experiments

There is a lot of churn in this PR, so to mitigate some of that, a large bulk of work was placed behind the nuget_native_analysis flag, but not all.

Changes that this PR incorporates whether or not the experiment is enabled:

• Workspace discovery (i.e., through file_parser.rb) has been updated to always report the full path of dependency files (e.g., /src/Directory.Build.props instead of just Directory.Build.props. This is required to avoid confusion during a multi-directory update, because it's possible for two directories to be given, src/ and test/, and if both have a Directory.Build.props file, later update stages don't know which file is being referenced. This change is reflected in the smoke tests.
• Metadata finder has been scaled back to barebones behavior. Workspace discovery, mentioned above, now reports the package info URL directly, instead of leaving it up to the metadata finder to manually poke the NuGet APIs to find a .nuspec file and then extract the repo. This change is reflected in the smoke tests.

Changes placed behind the experiments flag:

• The file update_checker.rb has been modified to check for the experiment flag and dispatch to the new update checker in native_update_checker/native_update_checker.rb when appropriate. If the experiment flag is not set, the old behavior remains. To disambiguate files and types with similar names, I've prepended native_ or Native to the new ones.

I have tried to keep the changes that check the experiments flag, move the new update checker, and apply the prefix Native... to a single commit so that when the experiment is over, that single commit can essentially be reverted (modulo some minor cleanup) to enable the new behavior.

I have verified the old and new behavior is invoked as appropriate by running an update job both with and without the experiments flag in the job.yml file.

---

I timed some update operations and the new behavior appears to be within a few seconds of the old/existing behavior with quite an overlap, so the end result is probably a wash.

Existing updater, 2 runs: 5:23 and 6:16
New updater, 2 runs: 5:51, 6:29

[jeffwidman]

I mostly skimmed it, and trusting your team to own this section of code. What parts I did poke at deeper looked good to me.

I have tried to keep the changes that check the experiments flag, move the new update checker, and apply the prefix Native... to a single commit so that when the experiment is over, that single commit can essentially be reverted (modulo some minor cleanup) to enable the new behavior.

😍

[jeffwidman]

nice! Making life simple down the road.

[brettfo]

I've had to re-figure this out too many times.

[jeffwidman]

Merging via a merge commit to preserve history... I find the "fix lint" commits ugly, but I don't want to squash down and lose the niceness where you'd put most of the feature-flag changes behind a single commit.