Add extension for ROM transactions

[waiting-for-dev]

We add a Dry::Operation::Extensions::ROM module that, when included, gives access to a #transaction method. This method wraps the yielded steps in a ROM transaction, rolling back in case one of them returns a failure.

We lean on a new Dry::Operation#intercepting_failure method, which allows running a callback before the failure is re-thrown again to be managed by the wrapping #steps call. Besides providing clarity, this method will be reused by future extensions.

The extension expects the including class to define a #rom method giving access to the ROM container.

class MyOperation < Dry::Operation
  include Dry::Operation::Extensions::ROM

  attr_reader :rom

  def initialize(rom:)
    @rom = rom
  end

  def call(input)
    attrs = step validate(input)
    user = transaction do
      new_user = step persist(attrs)
      step assign_initial_role(new_user)
      new_user
    end
    step notify(user)
    user
  end

  # ...
end

The extension uses the :default gateway by default, but it can be changed both at include time with include Dry::Operation::Extensions::ROM[gateway: :my_gateway], and at runtime with #transaction(gateway: :my_gateway).

This commit also establishes the dry-operation's convention for database transactions. Instead of wrapping the whole flow, we require the user to be conscious of the transaction boundaries (not including, e.g., external requests or notifications). That encourages using individual operations when thinking about composition instead of the whole flow.

[waiting-for-dev]

A quick note about being able to provide options to the #transaction method. It looks like when you invoke #transaction on a ROM relation, you can provide different options that are forwarded to the underlying engine. However, that doesn't look to be the case when you invoke Gateway#transaction (as we need to do here). ROM::Gateway will invoke a #transaction_runner method defined in every adapter and will forward options to it. For rom-sql, options are ignored.

@timriley, @solnic, please, just tell me if I'm missing something.

[timriley]

Looks good! Just a couple very minor things here but otherwise let's get this in :)

Let's make sure we get the issue with the #transaction arguments properly looked at, too, but AFAICT that shouldn't prevent this from being merged in the meantime?