notDeepLoosEqual to set .notExpected property instead of .expected

[hville]

Unlike t.notEqual and t.notDeepEqual, t.notDeepLooseEqual sets the reference value in the .expected property instead of the .notExpected property.

[ljharb]

This would be a breaking change, and it has no tests.

If this change isn't causing any tests to fail, tests will need to be added - otherwise, they'd need to be modified.

[hville]

The PR passes all of the current tests and is consistent with all other methods.

Creating a new test for this would invlove also creating new tests for deepEqual, notDeepEqual, deepLooseEqual and notDeepLooseEqual since the .expected and .notExpected properties are not validated in any of the current tests. This would be a seperate PR with a much wider scope.

[ljharb]

It's not solely up to me, but I'd rather refuse to merge a PR that doesn't come with at least a regression test. Even a single test case, that breaks without your change, would be a start. If tests are passing without your change - and your change is necessary - then the lack of a failing test is part of the same bug.

[hville]

Ok. Let me think about it because I don't really know what to test apart from the property itself. (I'm using these properties to pretty-format the output from the native result object instead of the text output)

[Raynos]

lgtm.

[hville]

Checked the TAP spec and node's assert to compare and devise relevant tests and found that node's assert only uses the .expected property for all assertion types instead of .expected and .notExpected. For consistency I would be better to change tape's properties accordingly (the opposite of the fix proposed in this PR) which could be a breaking change. I'm now using assert directly instead of jugling arround with tape's stream so this PR can be dropped and closed.

[ljharb]

This seems subsumed by #454.