This helps with writing and running bug reproducers using testscript.
Since they are often meant to represent the desired behavior,
and not the current behavior, they are designed to fail.

However, some reproducers consist of multiple commands,
and cmd/testscript would stop at the first command to error.
When running a reproducer which is designed to fail,
we want to run the entire script and see all failures.

Add a `-continue` flag which does just that.
With it, `T.FailNow` still marks the test as failed,
but does not stop its execution via a panic.
We also make two small changes to behave correctly when we reach the end
of a script but `T.Failed` is true; we should "FAIL" rather than "PASS".
The previous code only reached that point if no errors had happened.

Note that we also altered `runT` to use pointer method receivers.
It used atomics to modify and read its `failed` field,
but without pointer receivers, `T.Failed` always returned false.
It appears that bug had been present for a long time.

Co-Authored-By: Paul Jolly <[email protected]>