Improve the trace garbage collector

[pbiggar]

This adds a number of initial fixes to the trace garbage collector. I'm not sure it really worked at all, and it certainly didn't support wildcards, 404s, paths with a huge number of stored events (eg 2 million in one case) and maybe it mixed up path/modifier as well (not sure). This:

• combines the two garbage collectors which were the same code
• removes some old garbage collection code
• when loading 404s in the client, don't load the canvas data
• fixes bugs that were due to mixing up path and modifier
• adds ability to delete 404s
• adds support for GCing HTTP paths with wildcards (plus index to make it fast)
• reduces size of limit to 1000, but will repeat 100 times so long as there are 1000 deleted each time

[pbiggar]

Before merging, I should:

• add tests for 404s
• add a test for wildcards
• check that the select distinct on has an index.

Note that I've already added the index in production.

[pbiggar]

the select distinct on is from Stored_events.list_events, which is used to generate the 404 list. On our biggest user, generating that take 300 seconds. If we drop the trace_id and the timestamp, it's fully indexed and takes 15 seconds. The traceID is used when converting a 404 to a handler, which is annoying.

I'm experimenting with adding an index. Since it's used in 404s, we can't just get rid of it.