Don't store importer errors in the database

[RickMohr]

• Create table of error messages by error code
• append_error saves only error code, not error message
• Add errors_by_array_with_messages for situations where messages are needed
• Fix a few tests
• No data migration needed because error fields are accessed by name, not position

[steventlamb]

(pulled down the branch and did some grepping)

With this change, errors_as_array is used almost entirely inside models/base.py. There is one exception:
https://github.com/RickMohr/OTM2/blob/bf06f682b7a8071c8fe53c41a719430cd86917ff/opentreemap/importer/views.py#L534

It was probably always a good idea for this mutation logic to be factored into a model method, but it makes even more sense now. Consider factoring this out to a model method like GenericImportRow.strip_errors_by_code and making errors_as_array a "private" method, ie _errors_as_array. The more encapsulation of error processing, the easier it is to reason about.

[steventlamb]

consider e -> _e

[RickMohr]

I did try that, and thought it hurt readability. So I'm leaving it even though _e is more correct.

[steventlamb]

On initial review, this looks like a nice, clean way to handle this. I'll +1 after I run the branch and look at some db rows.

[RickMohr]

@steventlamb, other suggestions addressed.

[steventlamb]

+1 I tested that this is reading and writing error messages correctly. This is as expected for new records. For existing records, I took a tree import row with errors on three fields that was already in my database, it had msg attributes for these errors. I corrected one error and inspected the DB and the msg attributes were gone. Excellent.

[coveralls]

Coverage decreased (-0.04%) when pulling cc8bc80 on RickMohr:no-error-msgs-in-db into ba59ccb on OpenTreeMap:master.

[coveralls]

Coverage decreased (-0.04%) when pulling cc8bc80 on RickMohr:no-error-msgs-in-db into ba59ccb on OpenTreeMap:master.