Report server

[Ryan McNally]

While I really like the ability to browse reports straight from the filesystem:

• I suspect we're swimming against the current - we're already having to do make dubious edits to compiled artifacts to support it. I fear some future update will break things in a more unfixable manner.
• There's some really nice report functionality (interaction diagram, model diff tool, basis comparison) that only works with a properly-hosted report

It'd probably be a good idea to add some facility that would host execution reports that have been generated on the local machine so that the test can open a localhost url rather than the file directly:

• It'd have to be a separate process to the test - we want the test run to finish properly and then still be hosting the report after that
• It should shut itself down when the reports are no longer being viewed - reports would have to send some sort of heartbeat back to the hosting tool to keep it alive
• We'd only need a single instance hosting reports from multiple runs. How would a test run discover the instance? Multicast? Known port? Known file?
• It'd need some sort of gui: list of hosted reports, appear in the system tray, etc
• Ability to pin reports (they get copied to some temp dir) so that multiple runs of a single suite (that might otherwise overwrite reports) can be compared