Improve efficiency of GitHub querying

[agrski]

The current implementation fetches file metadata (path, binary/text status, file-type suffix, etc.) along with the contents of files.

For small files and/or when most or all files match the file-type filter(s), this does not impose any particular burden.
However, for other cases it is inefficient to fetch a lot of data just to throw it away immediately.
Instead, the querying logic should pull file metadata, determine which files it is interested in, and only then retrieve the file contents for matching against.

This should significantly reduce network bandwidth consumption and CPU usage, at the cost of potentially higher latency, at least until the first result is available.