Top level `frames_written` delayed update after `writing`

We've noticed that an individual `fp` (i.e., `fp0`) reads `frames_written==0` immediately when it goes into `writing==true`. However, if using the top level `fp` attributes, there is sometimes a large delay (e.g., `0.5s`) from `writing==true` to `frames_written==0`, which results in a misconfiguration in the kickoff context on `ophyd-async`.

## Steps To Reproduce
Steps to reproduce the behavior:
1. Use dodal's `configure_arm_trigger_and_disarm_detector` plan
2. See occasional timeout error on `bps.complete(..)`, as we erronously expect an extra exposure to come.

## Acceptance Criteria
- Race condition between top level writing and frames_written rectified, and this fix (TODO: link to code line) removed from `ophyd_async`