Dwgdbserver

- Added: disabled "run timers when stopped" in warmStart to make sure that this indeed the default.
- Restructured project structure to make it an installable packet
- Changed: Version number is now retrieved from importlib.metadata.version

- Changed: dw-harvest.py moved to deviceinfo/harvest.py
- Added: collect.py script in deviceinfo, which collects which MCUs are supported and what is the device ID/name mapping.
- Added: Early check in main whether device is supported
- Changed: We now use names instead of signatures in the error messages about the wrong MCU.

- Changed: Signals are now numbers, not strings anymore
- Changed: Instead of sending S-packets (containing only signal numbers), we now send T-packets with SREG, SP, and PC. Supposedly, this speeds up the time after a stop.

- Added: `monitor timers [freeze|run]` command implemented.

- Fixed: 'Address in use' errors happened regularly when starting dw-gdbserver after fatal errors. This occurs when the gdbserver closes the IP connection first. To avoid this, fatal errors are now caught on the level of monitor commands or the general handler, and the server is not terminated. This forces the user to terminate the debugger, after which the gdbserver can wait some time and then terminate, freeing the IP port immediately (see https://hea-www.harvard.edu/~fine/Tech/addrinuse.html)

- Fixed: Attaching to the AVR core now, because the XPLAINED-Mini board requested it.
- Fixed: XPLAINED-Mini only accepts full-page reads. Now, we either read from the cache or fall back to page-wise reading if there is nothing there.
- Fixed the problem of breakpoints not being removed when finishing.
- Implemented lazy breakpoint handling.
- Implemented assignment of hardware BPs to the most recently introduced breakpoints.
- Implemented two-word instruction simulation when starting at a breakpoint
- Implemented interrupt-safe single-stepping (+ the monitor function to enable that: `monitor singlestep safe`)
- Implemented monitor commands to disable fast loading and flash caching: `monitor load writeonly` and `monitor flashcache off` . Default is `readbeforewrite` and on, respectively.
- Implemented monitor commands for internal testing purposes: `monitor LiveTests` will initiate a live test of the gdbserver. Here the full command has to written out!
- Implemented monitor command to selectively switch new execution, single -stepping, and breakpoint handling on and off: `monitor Execution [old|new]`. Default is `old`.
- Renamed classes and modules to x... instead of dwe_ and X... instead of DWE. X standing for 'extended'.
- Tried out the SNAP, and it works without a hitch, albeit a bit slower than Atmel-ICE. Will measure load times with both.

- Fixed: Learned from xedbg that there exists a field called `buffers_per_flash_block` , which can be used to tell the debugger that the MCU has a 4-page erase command. With that, ATtiny1634, 841, and 441 work now as well!
- Changed: The property `buffers_per_flash_page` as found in the ATDF files is taken as it is and not changed to something else.

- Refactoring: New `MonitorCommand` class, which does all the bookkeeping and messaging
- Fixed: In the 'S' and 'C' commands, the signal was interpreted as an address

- Changed: Cleaned up info messages when starting up
- Changed: Streamlined debug log messages in GDB handler
- Changed: `continue` and `single step` now receive SIGHUP when not connected to debugWIRE
- Added: S and C record handling is now necessary because GDB propagates signals other than SIGTRAP
- Added: Check for overlapping chunks when flashing
- Fixed: Reading single flash bytes (even at mis-aligned byte addresses) works now.