Fix the mutation tester for a range overlap check

An assistant's test suite for a date-range overlap library reports 97% line coverage, all green. Coverage shows the lines ran, not that a test would notice a wrong answer. So the team built a small mutation tester. It plants small faults in the overlap check, called mutants, and reports the ones the suite fails to notice.

An assistant wrote surviving_mutants(cases). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• cases: the suite's inputs, a list of rows. A full row is [a_start, a_end, b_start, b_end]: two ranges of integers. The list can be empty.
• Return the names of the mutants that survive, as a list of strings.

Rules the tests check:

• Both ends of a range are included, so the range 1 to 3 holds 1, 2 and 3.
• Skip any row with fewer than 4 numbers. In a longer row, ignore every number after the fourth.
• The correct check: the ranges overlap when a_start <= b_end and b_start <= a_end. Apply it exactly as written, even when a range's start is after its end.
• The six mutants, each a wrong version of that check:
  • always_false: always false.
  • always_true: always true.
  • off_by_one: a_start <= b_end + 1 and b_start <= a_end + 1.
  • strict_left: a_start < b_end and b_start <= a_end.
  • strict_right: a_start <= b_end and b_start < a_end.
  • swap_and_or: a_start <= b_end or b_start <= a_end.
• A mutant is killed when at least one row gives a different answer under the mutant than under the correct check. Otherwise it survives.
• Return the survivors in alphabetical order, which is the order listed above. With no usable rows, all six survive.