Since #83 the loop commits the device layer once per frame (input_drive), but a
game that ALSO called Input.poll by hand committed a second time in the same
frame; input_device_commit copies in_held into in_prev at the top of every
commit, so the second commit left in_prev == in_held and the edges (held &&
!prev) could never see a transition.
Fix: an in_have_frame_driver flag. The loop's input_drive sets it; a manual
Input.poll under the loop then becomes a no-op returning the frame's key
instead of re-committing. An entry-driven harness has no loop, so the flag
stays false and each Input.poll commits a frame as before (the #7/#50
record/replay + device tests are unchanged). Frame loop now calls input_drive.
Example input_edge (press edge on the down frame, release edge on the up
frame). Full suite 110/0, goldens byte-identical, fixpoint holds.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
