Report schedule_all start errors

[fallintoplace]

Summary

Fix exec::schedule_all on static_thread_pool so exceptions thrown while preparing item operations are reported with set_error(std::exception_ptr) instead of escaping from start().

set_next is allowed to throw, and schedule_all advertises set_error(std::exception_ptr), but the previous start() noexcept implementation called set_next and connect without a catch boundary. The operation now constructs all child operations before starting any of them, catches preparation failures, and sends the current exception to the receiver. It also tracks how many child operations were actually constructed so partially built operations are cleaned up correctly.

Testing

• cmake --build build --target test.exec
• ./build/test/exec/test.exec "schedule_all on static_thread_pool sends errors from set_next"
• ./build/test/exec/test.exec "[types][static_thread_pool]"
• ASAN_OPTIONS=abort_on_error=0:symbolize=1 ./build-asan/test/exec/test.exec "schedule_all on static_thread_pool sends errors from set_next"
• ./build/test/exec/test.exec

[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.