[LIT] Migrate runtime lit config to the internal shell (#24794)

Part of #24607.

This switches `runtime/lit.cfg.py` to the internal shell and removes the
`execute_external` / `force_execute_external` workaround. It's the first
of the seven configs listed in the issue. I'm doing one at a time so the
failures stay easy to review.

The internal shell has no subshells, so the `( ... )` wrappers in
`flags_test.txt` were read as a command named `(` and the test died with
exit 127. Two kinds of RUN lines needed fixing:

- **26 lines** wrapped a single command in `( ... )` before piping to
FileCheck. The subshell did nothing there, so I dropped the parentheses.
- **3 lines** used `( cmd 2>&1 || [[ $? == 1 ]] )` to tolerate an
expected non-zero exit. Those now use `not`, which is the usual spelling
for expected-failure RUN lines and also asserts the command really does
fail. `not` is declared in the suite's tools, the same way
`tools/test/BUILD.bazel` does it.

Same commands and same FileCheck prefixes, so no coverage is lost.

## Testing


ctest -R
"flags_test.txt.test|modules/check/test/(failure|success|unavailable).mlir.test

All 4 tests pass on the internal shell, matching the baseline before the
change. The three modules/check tests are plain pipelines and needed no
edits.

One thing I checked: rewriting those lines could have made them stop
running while still going green. I broke an expected UNKNOWN-FLAG
pattern on purpose, confirmed the test failed, then put it back.

Verified on macOS/arm64 only - relying on CI for Linux and Windows.

Assisted by: Claude Code

Signed-off-by: Zohaib Shah <zohaib.shah.2047@gmail.com>
Co-authored-by: Zohaib Shah <zohaib.shah.2047@gmail.com>