This test fails with the error:

    AssertionError: Command 3 (dmypy check -- bar.py) did not give expected output
    --- Captured stderr call ---
    Expected:
      bar.py:2: error: "type: ignore" comment without error code  [ignore-without-code] (diff)
      == Return code: 1                             (diff)
    Actual:
      (empty)

This test illustrates that '"type: ignore" comment without error code'
errors currently disappear in the same way that 'Unused "type: ignore"'
errors do as described in python#9655.

Ref: python#9655

This test fails with the error:

    AssertionError: Command 3 (dmypy check -- bar.py) did not give expected output
    --- Captured stderr call ---
    Expected:
      bar.py:4: error: Name "a" may be undefined  [possibly-undefined] (diff)
      == Return code: 1                             (diff)
    Actual:
      (empty)

This test illustrates that possibly-undefined errors currently disappear
in the same way that 'Unused "type: ignore"' errors do as described in python#9655.

Ref: python#9655

These tests show how some errors disappear on a re-run of dmypy after a
file is altered.

This which fixes issue
python#9655 wherein some types of error
would be lost when a file was re-processed by dmypy. Regression tests
are also included.

This also fixes another error where sometimes files would not be
re-processed by dmypy if the only error in the file was either "unused
type ignore" or "ignore without code".

This catches a regression caused by the previous commits where "type:
ignore" comments are erroneously marked as unused in re-runs of dmypy.

As far as I can tell, this only happens in modules which contain an
import that we don't know how to type (such as a module which does not
exist), and a submodule which is unused.

Before this change, fine-grained builds could spuriously mark ignored
legitimate errors as "unused ignores".

By keeping track of these ignored errors we ensure that enough analysis
is done to know that the ignored lines are actually useful.

We have to change is_errors_for_file so that we don't consider files as
faulty when all of their errors were hidden.