I pushed a new, much simpler approach. Now, use environment variables to indicate when a download has already occurred. This is preferable because it will apply to any sub-processes as well (e.g. any brew install that we trigger within Homebrew, build.rb, postinstall.rb, test.rb)

Also, to clarify why mtime and --time-cond are insufficient here: these checks simply ensure that enough time has passed in between fetches, and that a new fetch will actually yield a new file. However, when building from source, it's possible that an installation will take long enough that new data is available and enough time has passed that Homebrew::API considers the file to be stale. This means that future calls to API data will try to download the data once again, which is what this PR is attempting to prevent. I've updated the description to make this clearer.

Note: I haven't tested this thoroughly yet on my machine, but I will before merging

Thanks for the help and iteration here @MikeMcQuaid!