Support for creating unlogged tables (simplified)

[rwillians]

This PR introduces the :modifiers field to Ecto.Migration.Table struct, which can be provided as an option to Ecto.Migration.table/2.

Modifiers¹ are known keywords that can be inserted between the tokens "CREATE" and "TABLE", for example the "UNLOGGED" keyword in PostgreSQL (see Create Table - PostgreSQL).

create table(:sessions, modifiers: "UNLOGGED") do
  # ...
end

The statement above now produces the following DDL:

CREATE UNLOGGED TABLE sessions ...

Also, the DDL logs have been updated to include the modifiers when creating a table (seemed like a nice touch):

$ mix ecto.migrate

17:11:49.826 [info] == Running 20251128173650 MyApp.Repo.Migrations.CreateWebhooksTable.change/0 forward

17:11:49.828 [info] create unlogged table webhooks

17:11:49.838 [info] == Migrated 20251128173650 in 0.0s

---

¹ It was discussed here the use of the nomenclature parameters for this, but parameters aren't limited to show up between "CREATE" and "TABLE". For example, they can appear at the end of create table the statement. That's why I'm using the term modifiers.

[greg-rychlewski]

Thanks for the PR. It looks good. Besides the comment I left we should also add this to MySQL and SQL Server or raise if they don't support it.

[greg-rychlewski]

Suggested change

	defp modifiers_expr(<<_, _::binary>> = value), do: [String.upcase(String.trim(value)), ?\s]
	defp modifiers_expr(value), do: error!(nil, "PostgreSQL adapter expects :modifiers to be a non-empty string or nil, got #{inspect(value)}")
	defp modifiers_expr(modifiers) when is_binary(modifiers), do: [modifiers, ?\s]
	defp modifiers_expr(other), do: error!(nil, "PostgreSQL adapter expects :modifiers to be a string, got #{inspect(other)}")

This is for consistency with similar things like options. The existing pattern is not to normalize and not to care if they send an empty string. Raising seems a bit harsh for empty string as well since it's still valid SQL and they might populate this from a configuration value that is empty in some environments.

[rwillians]

Makes sense, thanks for explaining

[rwillians]

Thanks for the review @greg-rychlewski, I'll most likely have an update on it in the next weekend