Use model timestamps when available

[andrepiske]

It changes how the value for created_at column in irontrail_changes rows is determined. Originally, it was just set to whatever postgres's NOW() function would return. This changes it so that it considers the models' timestamp attributes (created_at and updated_at). Note: it doesn't account for *_on or other name variations for timestamps.

The logic is the following:

1. When inserting a record, it tries to use the value of the created_at column. Otherwise*, it uses the value of NOW().
2. When updating a record, it tries to use the value of the updated_at column. Otherwise*, it uses the value of NOW().
3. When deleting a record, it just uses the value of NOW().

---

Related tasks:

• INF-234

This is an alternative to #17 -- there could be a hybrid of both, but I feel for now let's take the simpler route of this PR only. The only "odd" case is that it would be harder to determine the creation date of irontrail_changes records in a test environment, since there's no way to mock it. But I understand that the way it is covers most uses cases with less overall complexity, while still preserving a predictable and good behavior in production environments.

[izn]

LGTM! 💯