Bundle list routing + other bugfixes

[vdye]

Fixes #32

While end-to-end testing, I ran into #32; while investigating that bug, I found a couple others. This PR fixes all of the ones I've found, and adds some unit test coverage for good measure.

• Commit 1 stops the trace2 TraceLogger from suppressing the message intended to be logged to the console in Fatal() & Fatalf().
• Commit 2 fixes a condition I had inverted; it was intended to prevent the trace2 logger from logging an "error" event in every logger.Error*() call up the call stack (and only log the first it sees), but instead it logged every error except the first one. 🤦‍♀️
• Commits 3-4 do some refactoring to prepare for unit testing GitHelper
• Commit 5 fixes an issue in GitHelper.CreateIncrementalBundle(). Specifically, the function was intended to return no error if the Git error message indicated that any new incremental bundle would be empty; that message wasn't captured, so CreateIncrementalBundle() would never see it and always return an error.
• Commits 6-9 refactor BundleProvider.WriteBundleList to write two bundle lists: one with bundle paths relative to a request URI with a trailing slash (with relative paths like bundle-123456.bundle), and one (new) for URIs with no trailing slash (relative path like myrepo/bundle-123456.bundle). This is the first step to fixing how we serve bundle list requests with no trailing slash.
• Commits 10-11 refactor bundleServer.serve() to serve the relative-to-org (rather than relative-to-repo) bundle file if there is no trailing slash on the request URI.
• Commit 12 adds a step to first fetch from the remote when creating an incremental bundle in git-bundle-server update. Without that, no new incremental bundles will ever be generated.
• Commit 13 adds the bundle.heuristic config to the bundle lists served by the bundle server, ensuring incremental bundles will be applied on fetches after an initial clone.

Other approaches to #32

I also considered generating the bundle list config contents on the fly (by reading the bundle-list.json and formatting the output), but the cost tradeoff of computation for every request vs. a bit of extra disk space seemed to favor the latter.

[vdye]

Upon doing more testing, I've found more bugs (e.g., git-bundle-server update never calls git fetch, so no new bundles will ever be created). I don't think that affects the other bugfixes (those can be reviewed as-is), but I will be pushing more things to this branch in the near future.