remove indexed types

[tpolecat]

This is big PR, sorry, but relatively straightforward.

This removes indexed types (and indeed all parameterization in the main model) in favor of simpler (but more verbose) ADTs with cases per instrument.

• Program is now a simple case class.
• Observation is now an ADT with constructors for each instrument. So Observation.GmosN is a GMOS North observation and all its data members are specialized to GMOS North.
• TargetEnvironment is now an ADT with constructors for each instrument.
• Asterism is now an ADT with constructors for each asterism type (one per instrument now, but there will be more).
• StaticConfig and DynamicConfig are no longer indexed on the singleton type of the instrument.
• Step is now an ADT with constructors for each instrument, and the sense is reversed with respect to the "base" step types like Bias and Science, which are now members of each constructor rather than the reverse. This lets us say things like type GmosSequence = List[Step.GmosN] for instance.
• The Self type member and matching Aux pattern which were added to support singleton indexing of gem.enum types are no longer necessary and have been removed.

The data layer is largely the same, with a few exceptions:

• Asterisms are now fetched by joining with observation and no longer need the AsterismType hint. This simplifies things downstream but means the query may become complex as more asterism types are added. But I think it's the right way to do it.
• Queries that returned "degenerate" versions of observations and programs now return tuples. We may want to add more structure but I think this will depend on usage.

Some misc. cleanup is technically unrelated but aided in the refactoring.

• SmartGcal is now in its own source file under gem.config
• Instrument names have been normalized based on the Instrument enumeration. Uses of F2, GmosNorth, and GmosSouth have been normalized to Flamingos2, GmosN, and GmosS, respectively so everything matches up.
• Cats instances have been added for TreeMap and TreeSet, by copy/pasting the existing instances for SortedMap and SortedSet, respectively.

Open questions, to be addressed later:

• Sequences are List[Step] in the model but remain as TreeMap[Location.Middle, Step] in most of the data access layer. We may want to push the map representation all the way up.
• Uses of TreeSet and TreeMap should probably become SortedMap and SortedSet, respectively, if possible, to take advantage of existing instances provided by Cats.
• Adding a new instrument is a lot of work and this makes it harder because there are new constructors and new pattern-match clauses to add. But on the whole I think the simplification makes the code a lot more straightforward.

Followup tasks:

• The Json stuff needs to be re-organized and set up to use Circe's semiautomatic derivation, which requires you to declare all the instances but ends up being much faster to compile. I can circle back and do this.
• Gnirs isn't on the list of observation types we generate for our tests, so I added it but it caused some test failures so I removed it again. So we need to track that down.
• The Instrument enum (and all pattern blocks derived from it) are unordered. Let's sort them by instrument name.

[tpolecat]

We should probably squash-merge this because nothing prior to the last commit will pass CI.

[tpolecat]

Let me know if any of my responses alarmed you. If not I'll merge.

[swalker2m]

Let me know if any of my responses alarmed you. If not I'll merge.

No alarms and no surprises. Merge away. 👍