Remove fixed max reader number

This is a constant footgun for fork maintainers, so we can assume it's even more so for people trying to casually hack on openpliot or msgq. https://github.com/commaai/msgq/pull/681

An immediate idea is to split the readers out into a separate file on `/dev/shm/`.

Start by reading linked pull request #681 and inspect the reader bookkeeping and shared-memory layout in the msgq repository. Evaluate the proposed separate reader file under /dev/shm/, then confirm that reader counts are no longer constrained by a fixed maximum and that existing pub/sub behavior remains intact.

Written by the indexing model from the issue text.