Add custom knex snake case mappers

https://eaflood.atlassian.net/browse/WATER-3896

Whilst working on [Add SROC Supplementary Billing Invoice Service](#119) we hit an issue. It was the first time we needed to query the `crm_v2` schema. But that threw errors due to the way [Objection.js knexSnakeCaseMappers()](https://vincit.github.io/objection.js/api/objection/#knexsnakecasemappers) works. It uses Knex's [wrapIdentifier()](https://knexjs.org/guide/#wrapidentifier) and [postProcessResponse()](https://knexjs.org/guide/#postprocessresponse) hooks to see each 'identifier' name. It can then test whether it needs converting, either from or to snake case.

For example, `knex('table').withSchema('foo').select('table.field as otherName').where('id', 1)` will call `wrapIdentifier()` for the values `'table'`, `'foo'`, `'table'`, `'field'`, `'otherName'` and `'id'`.

**knexSnakeCaseMappers()** takes some options, one of them being `underscoreBeforeDigits`. If we didn't use this fields like `address_line_1` or `section_127_agreement` in the DB would be incorrectly converted to `address_line1` and `section127_agreement`. But this is what leads to our problem.

The previous teams' decision to name one of the schemas `crm_v2` leads to an incorrect conversion. **knexSnakeCaseMappers()** is seeing this and returning `'crm_v_2'`. It doesn't know it's a schema instead of a column name because Knex doesn't provide that context. This is the exact same issue we faced in [Make timestamps consistent at model layer](#85).

Thankfully, a solution that didn't work there will work here. We can provide our own custom knex snake case mappers implementation which knows to ignore `'crm_v_2'`, whilst calling Objection.js own methods for everything else. It also gives us a solution if we face any more funnies like this when dealing with the legacy data.