fix: Test Manifest, parameterized do not override sibling failures - #15831

Fixes elixir-lang#15820

Alternative approach to elixir-lang#15825.

In short, parameterized tests aen't accounted for in the test manifest,
so if a variant first failed it could get overridden by a later run
with different parameters as the key is just {module, test}.

Based on [feedback](elixir-lang#15825 (comment)),
this PR tries to solve this by keeping _all_ parameterized tests marked
as failing.
So for variants 1, 2, 3, 4 ran and only 2 failed, all 4 get rerun which isn't
too much of a cost.
For this small cost, we can keep the format of the manifest file,
as it was before.

There is one wrinkle to this, which is I couldn't find a good way
to stop the override ("do not delete a failing test from the manifest,
if I have the same id") while still ever deleting a failing test
(which we need, otherwise `mix test --failed` would be ever growing)
other than to track which tests failed _this run_. So, we only
refuse to clear it if we know the test failed _this run_.

The additional work to track is quite minimal though.

It adds an additional argument to `put_test` though, which I
decided to give a default to not increase the splash radius of the
PR. Worth potentially removing, unless we think it's fair game
for external callers.