Destroy parallel bulk predecessor values

[fallintoplace]

Summary

This fixes the parallel scheduler bulk receiver so predecessor values stored for bulk execution have a real lifetime and are destroyed on every terminal path. The old implementation placement-newed a tuple into raw storage and even noted that the tuple was never destroyed, which leaks resources for owning predecessor values.

The storage now uses __manual_lifetime, destroys the stored tuple before forwarding set_value, set_error, or set_stopped, and adds coverage for value, error, and stopped completion paths. The test uses an owning value that tracks live instances so the old behavior leaves the count non-zero.

While adding coverage for value-carrying predecessor senders, the same path also needed a few adjacent fixes that were previously not instantiated by tests:

• expose the value completion domain from parallel_scheduler
• match bulk_item_receiver_proxy::execute(size_t, size_t)
• pass (count, receiver, scratch) to backend schedule_bulk_*
• invoke the connect initializer as an rvalue

History check: the raw-storage tuple and non-destruction comment were introduced in b67a1e58 when parallel_scheduler moved into the STDEXEC namespace, then persisted through later formatting/refactors.

Testing

• cmake --build build-parallel-scheduler --target test.stdexec -j8
• ./build-parallel-scheduler/test/test.stdexec "bulk on parallel_scheduler destroys stored predecessor values*"
• ./build-parallel-scheduler/test/test.stdexec "[scheduler][parallel_scheduler]"
• cmake --build build-parallel-scheduler --target test.parallel_scheduler_replacement -j8
• ./build-parallel-scheduler/test/test.parallel_scheduler_replacement
• git diff --check upstream/main...HEAD

Not run locally: clang-format --dry-run --Werror include/stdexec/__detail/__parallel_scheduler.hpp test/stdexec/schedulers/test_parallel_scheduler.cpp, because clang-format is not installed on this machine.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.