[AIRFLOW-4961] Set TaskFail.duration as int match DB schema column type

[fahran]

When writing a task failure record, convert 'duration' decimal
value -> an int before persistence, to remove reliance on the
database doing this automatically and gracefully.

---

Make sure you have checked all steps below.

Jira

• [✅] My PR addresses the following: https://issues.apache.org/jira/browse/AIRFLOW-4961

Description

• [✅] Here are some details about my PR, including screenshots of any UI changes:
  I raised https://issues.apache.org/jira/browse/AIRFLOW-4961 earlier today, to cover a minor issue we encountered whilst running airflow using CockroachDB. That ticket covers the full explanation of this PR.

Tests

• [ 🤔] My PR adds the following unit tests OR does not need testing for this extremely good reason:
  There aren't currently unit tests for testfail.py (that's not to say there shouldn't be), and an integration test would pass with the present code anyway, as Postgres' error handling masks the issue.
  It's also a pretty trivial change. Happy to add a test for testfail.py if we feel it's needed though.

Commits

• [ ✅] My commits all reference Jira issues in their subject lines, and I have squashed multiple commits if they address the same issue. In addition, my commits follow the guidelines from "How to write a good git commit message":
  1. Subject is separated from body by a blank line
  2. Subject is limited to 50 characters (not including Jira issue reference)
  3. Subject does not end with a period
  4. Subject uses the imperative mood ("add", not "adding")
  5. Body wraps at 72 characters
  6. Body explains "what" and "why", not "how"

Documentation

• [ ✅] In case of new functionality, my PR adds documentation that describes how to use it.
  • All the public functions and the classes in the PR contain docstrings that explain what it does
  • If you implement backwards incompatible changes, please leave a note in the Updating.md so we can assign it to a appropriate release

Code Quality

• [ ✅ ] Passes flake8

[ashb]

To replicate my comment from the Jira: should we change the column type instead? sub-second accuracy on the duration of an airflow task probably isn't all that useful is it?