Revert "Cherry-pick two upstream lit changes to optimize swift testing performance"

[edymtt]

Reverts #2687

This change will cause fallout on Swift bots running Ubuntu 16.04, which are running Python 3.5, which a failure similar to

 cd /Users/docker_user/src/build/buildbot_incremental/llbuild-linux-x86_64/tests && /usr/bin/python3.5 /Users/docker_user/src/llvm-project/llvm/utils/lit/lit.py -sv --param build_mode=Release /Users/docker_user/src/build/buildbot_incremental/llbuild-linux-x86_64/tests
Traceback (most recent call last):
  File "/Users/docker_user/src/llvm-project/llvm/utils/lit/lit.py", line 3, in <module>
    from lit.main import main
  File "/Users/docker_user/src/llvm-project/llvm/utils/lit/lit/main.py", line 13, in <module>
    import lit.cl_arguments
  File "/Users/docker_user/src/llvm-project/llvm/utils/lit/lit/cl_arguments.py", line 11, in <module>
    class TestOrder(enum.Enum):
  File "/Users/docker_user/src/llvm-project/llvm/utils/lit/lit/cl_arguments.py", line 12, in TestOrder
    DEFAULT = enum.auto()
AttributeError: module 'enum' has no attribute 'auto'

Addresses rdar://75786134

[edymtt]

@swift-ci please test

[edymtt]

This is also affecting PR testing, e.g. #2745

[davezarzycki]

I'd like to just fix the underlying problem and not revert the cherry picks wholesale. Can we please abandon this in favor of #2749? Thanks!