fix: keep fields of custom_index in format that they were provided

[vonagam]

Right now fields of custom_index are first transformed into atoms and then during migration generation into strings.

I used a field named "order" and encountered an error because of this. Ecto generates different statements depending on whenever a field was provided as a string or as an atom:

create index(:some_table, ["order"]) # invalid
create index(:some_table, [:order]) # valid

For atoms the resulting name will be in quotes, for strings they will be left as is (and it, for example, allows specifying things like "name gin_trgm_ops" for full-text search usage).

So it makes sense to keep user provided input as is and not convert it into atoms or strings.

Also made RemoveCustomIndex operation to use reverse direction of AddCustomIndex to avoid repetition.

[zachdaniel]

I think you're going to have to handle this when loading snapshots and encoding snapshots as well. Once we load the json snapshot, they will always end up being strings.

[vonagam]

Hm... makes sense. Can close then.

Alternative though maybe to always use atom form except when there is a white space involved.

[zachdaniel]

Oh, I think it's solvable, we just need to do a bit of extra work. When encoding we should store it as {"type": "atom" | "string", value: "value"}, and then decode it appropriately.

[zachdaniel]

And handle if its just a string for backwards compatibility.

[vonagam]

Ok, implemented such thing.

[zachdaniel]

awesome work! 🚀 Thank you for your contribution! 🚀