This should fix the segfault upon exiting of the Python interpreter when having
hxtorch and/or pynn-brainscales loaded (caused by a now-non-POD global variable
holding a logger instance — multiple destructions on the same instance happening).

This one includes a fix in hxtorch that wasn't caught when testing
locally (at UHEI) against newer pybind11 versions.

Previously, we used a custom do_fetch() to download some additional files
needed for the build and to prepare the build folder.
Now, we use spack to track all resources (i.e. download the additional files
itself) and only setup the build folder in install().

This reverts commit 674e26c0

This reverts commit 017e5e23

* a few commits in pynn-brainscales (2), grenade (2), halco (2), hxcomm
  (2), haldls (1), hwdb (1)
  * add interface-draft for plasticity rules
  * support for requestion server-side information
    (`get_bitfile_info` and `get_repo_state`)
* also cleans up the tag mess in the dependent repositories…
  really same tag everywhere
* same for oppulance; as software state didn't change, we don't require
  oppulance to match pynn-brainscales version (rc2 is sufficient)…
  relaxing constraint, rc2

If starting from scratch the archive file isn't there in do_fetch;
in this case, we can make use of the folder where the freshly
downloaded resources are cloned to.