- chore: remove unused headers

This creates rather annoying issues when used with a TransferGroup, in
which case the exception message is totally invisible ("terminate called
without an active exception").

A more permanent solution should execute checks when the formula is
compiled, so the logic_error can be thrown earlier.

Adapt to minor changes with clang-format-19

* feat: Implement tagging on registers
* Update include/RegisterInfo.h

Co-authored-by: Martin Hierholzer <martin.hierholzer@desy.de>

The implementation for partial accessors in SubdeviceBackend 2reg and 3reg are currently broken.
- Throw an exception if partial 2reg/3reg accessor is requested to inticate to the user that this is not working.
- Disable the according SubdeviceBackend unified tests

update project template

This allows us to use target_compile_features to set the C++ standard
in cppext and move to a newer standard.

The SharedDummy backend was crashing with a
boost::interprocess::lock_exception when a previous instance was
interrupted in the wrogn place. This is now fixed by catching the
exception and removing the shared memory segment (which is then broken
anyway).

Dietrich Rothe authored 2 months ago and Dietrich Rothe committed 2 months ago

* existing problem: even after board was removed/shut down, reading normal (not dma) pci registers
  does not throw an error from QtHardMon.
  To fix it, check state of device node after read and write of mmaped pci bar
* existing problem: when triggered by user interrupts, device error is never noticed, in particular when
  device does not send events anymore
  To fix it, add timer events for periodic devide node check
* existing problem: exceptions from event handler thread never caught
  To fix it, add try/catch around boost::asio processing loop

The XDMA backend called internally close() when a recovery of an
exception was attempted. In particular if the recovery was still
unsuccessful, the device was then left in the closed state leading to
logic_error exceptions in subsequent read/write operations. This caused
ApplicationCore applications to terminate immediately after an exception
of an XDMA device was seen.

This is now fixed by using a dedicated flag for the backend opened
status which is independent of the underlying interface status.

This is required to terminate ApplicationCore applications properly when
a poll-type input (connected internally) is waiting for initial values
(which uses a blocking call to future_queue::pop_wait()).