fix: don't strip valid characters from secret data

[james-garner-canonical]

Description

create_secret_data unnecessarily attempts to strip trailing = symbols from secret data, and inadvertently strips all trailing P, Q and = characters instead. This PR drops the unnecessary removal of trailing characters, and adds unit tests covering cases that would previously have been handled badly (as well as some test cases validating existing behaviour that wasn't covered).

Fixes: #1275

QA Steps

Unit test should pass. Locally I have tests/unit/test_jujudata.py::TestJujuData::test_verify_controller_uninitialized failing both on main and on this branch.

[james-garner-canonical]

Just changing to assert data ==, which results in dedenting the assertion data.

[james-garner-canonical]

Drop mostly unused unittest.TestCase base class so that pytest.mark.parametrize can work.

[dimaqq]

Code LGTM

let’s get clarity on failing integration tests vs main

[james-garner-canonical]

Manual run on main in progress, let's see how it goes.

[james-garner-canonical]

Summary: it's the same on main.

---

The non-quarantined integration test failures in this branch are all with Juju 3.3. #1267 noted multiple test failures in main with 3.3 specifically in May this year and #1269 proposed to drop 3.3 from without pushback (still unmerged due to ideas of dropping more). The same failures are present on main (and actually we get an extra failure on main in one case, which I suspect is just random flakiness rather than anything to do with this PR).

Quarantined integration tests are failing test_relate on all Juju versions other than 3.1. The jobs are still running on main, but that specific test has already failed (3.3, 3.4, 3.5, 3.6). I've opened #1277 to document the test_relate failure.

Additionally, several more quarantined integration tests fail on 3.3 specifically (3.4, 3.5, and 3.6 only fail test_relate). This is also the case on main.

[james-garner-canonical]

/merge