Reimplement Stackless/unittests/support.py. The new implementation
is more IDE friendly and requires only a single pass to run all
test with and without soft switching.

Base all tests on StacklessTestCase.
Remove a second call of unittest.main from ```if __name__ ==
"__main__"```-blocks

https://bitbucket.org/stackless-dev/stackless/issues/86
(grafted from 8237bba0d27ab1bdd2cdbfa52c7bdb6f468119a7, 719923b5c638,
06295530c672, 65cff13662f2)

See #86

This does not fix races where the _localimpl is delete by another
thread... but I don't think the _threading_local implementation
is actually used.

…ngly. Closes python#86.

Per pythia python#86 + supervisor [chat L2469]: bridge inventory is a
COMPLETENESS ASSERTION (this set of N bridges is sufficient). W25 §5.3
lesson: compile-clean enumeration is necessary but not sufficient —
inventory can have hidden gaps (another bridge silently absorbs drift).

Adds Step A falsification phase (~10min, MANDATORY before Step B):
- testkeeper mutates one bridge signature post-LITE-SPEC
- BUILD_EXIT=2 at expected call site = inventory correct
- BUILD_EXIT=0 = inventory gap, reassess before Step B

General principle for future workstream specs: Step A falsification
test required in gate-completion criteria. Bridge inventory
completeness must be empirically falsified, not just enumerated.

Also updates Step A bridge count 5→4 per generalist [chat L2461]
empirical scope reduction + theologian [chat L2466] (B) combined-bridge
decision.