add options to override config file and tiles file

[sbailey]

This PR adds several features useful for small scale functional testing:

• --config-file option for scripts enabling overriding the default configuration file
• new config file tiles_file parameter to specify the DESI tiles file to use (requires desimodel >=0.9.0; defaults to desi-tiles.fits as before)
• adds rules-layers.yaml with simplified "do layer N before layer N+1" rules for testing

Other changes along the way:

• rename output directory default environment variable $DESISURVEY -> $DESISURVEY_OUTPUT since $DESISURVEY was already in use for installed packages at NERSC
• prints ERROR message (but not FATAL crash) if any rules that don't cover any tiles
• fixes 0.0/0.0 rules bug when there is only one tile being evaluated on a single tile such that hi==lo
• tests work with desimodel test-0.9 branch of data instead of requiring trunk (on my laptop at least; we'll see if Travis agrees).

A companion update to surveysim will add the --config-file option there and update the tutorial instructions.

I suggest making tag 0.9.0 after this is merged (version file was already update but tag never made...)

[dkirkby]

I did make the 0.9.0 tag at the appropriate time, but forgot to push it (now fixed).

[dkirkby]

I see you edited changes.rst assuming that I didn't make the tag, so please update that.

[dkirkby]

Let's pull the docs at the head of rules.yaml and now rules-layers.yaml into a separate file (README.md?) so it isn't duplicated and easier to maintain.

[coveralls]

Coverage decreased (-0.1%) to 38.163% when pulling 40f3432791a2d57243f3febe1dd77912a2a3341a on config into 9674ae960e3f71d18e72a397c82c51e51b6f29ca on master.

[sbailey]

• Updated changes.rst with tag history
• Fixed lo==hi logic (last minute "clarity" refactor before pushing broke... good catch)
• Pulled rules.yaml and rules-layers.yaml doc header into separate doc/rules.rst doc.

[coveralls]

Coverage decreased (-0.2%) to 38.148% when pulling 4ef6ad70a7e98937b2a67ec13e53f4df5478830a on config into 9674ae960e3f71d18e72a397c82c51e51b6f29ca on master.

[coveralls]

Coverage increased (+18.6%) to 56.874% when pulling df848e8d9ba52ae939ba6753475e9fb36005156d on config into 9674ae960e3f71d18e72a397c82c51e51b6f29ca on master.

[sbailey]

postfacto comment: the new unit tests run fast when using a desimodel test data subset, but are too slow when running on desimodel svn trunk data with all tiles. I will fix that now.