DB fixes and Scheduler HWM

[minrk]

This branch is actually based on my 'windows' branch, so the real diff is here:
minrk/ipython@windows...db

until PR #374 is merged

Basic features:

add HWM to Python scheduler, allowing users to tune level of job queuing. (HWM=1 behaves like old TaskScheduler, HWM=0 is default, and behaves like pure 0MQ).

DB backend changes:

• add db.get_history() method (like client/view.history, but returns all known msg_ids)
• db.find_records returns list, allows filtering of keys
• add db_backend tests

Two new Client methods exposing db backends:

• rc.hub_history() # proxy for db.get_history
• rc.db_query() # proxy for db.find_records

Up for discussion: These are Client methods, not View methods. To me, they seem like low-level methods, and thus appropriately bound to the low-level object, but They can be easily propagated up to the Views.

Extra fixes along the way:

• Harder for db issues to crash the Hub
• Better JSON handling of date objects
• Don't raise NoEnginesRegistered when load-balancing on all engines
• rc.get_result works with whole history if database is persistent

[minrk]

Among the DB bugs fixed by this is a failure in MongoDB.add_record, if buffers were unspecified (and a method name typo), as reported in #391.

[kaazoo]

Requesting all known tasks and creating tasks in advance (when no engines are running) works now. Thanks.

[ellisonbg]

I agree that these should be Client level methods, not View level.

[ellisonbg]

As long as this is tested, I think it can be merged. I gave it quick look through and things look fine.

[minrk]

I've got db_backend tests, but I'll add them to the client methods before merging.

[ellisonbg]

OK great.