Address nested-translation errors + refactor

[seancribbs]

In an effort to discover the nested errors that can/do occur in Riak's schemas, this substantially refactors the code in cuttlefish_generator.

This PR targets develop but should also be backported/merged to 2.0 branch for the next patch release.

One source of the problem (as shown in the backtrace in basho/riak#590) was that "extended" datatypes that failed to parse halted all attempts to parse other datatypes in the list. This meant that when calling cuttlefish_generator:map/2, if something was already coerced to a valid datatype, it might fail to parse as the extended type and result in the nested configuration failing.

The main refactors are:

• 16a34ed : Broke cuttlefish_generator:apply_translations() up into smaller, more digestible functions with fewer responsibilities.
• 71f063c : Caught exceptions resulting from parsing standard datatypes such that they will not completely abort the process.
• cc78e53 : Added more guards to cuttlefish_datatypes:from_string() and to_string() that caused things like list_to_atom(36000) (as in the example in basho/riak#590).
• 039ca0a : Refactored cuttlefish_generator:transform_type() so that the functions it calls are not mutually recursive. The control flow was extracted out into a monadic combinator foldm_either, while the nested calls simply return {ok, Value} or {error, Reason}.
• df4c280 : Replaced all manual formatting of varialbes now that cuttlefish_variable:format/1 exists. Found and fixed a resulting bug in cuttlefish_escript:describe/2.

[hazen]

Now that my EQC is back online:

All 288 tests passed.

[hazen]

:+1: 2edaf38

[seancribbs]

@javajolt Even though it passed this time, I should probably fix it so it doesn't randomly fail again. Hold on.

[hazen]

Looks like repeated test runs are now passing:

All 288 tests passed.

[hazen]

:+1: 52041cd

[seancribbs]

@borshop merge