Promise launcher compares current slapgrid version with an old version
number to determine if some feature is supported or if it should use a
legacy workaround. But the comparison was a simple lexicographic order
string comparison, and now that we incremented from 1.9.3 to 1.10.0 it
wrongly compared the versions. Fix it with pkg_resources.parse_version.

When the range available to allocate a single IPv6 to each partition is
too small, wrap around by allocating the same IPs instead of aborting.

The wrap around logic loops over the addresses in the range other than
0 and 1 which are reserved for the slapnode, except if the range is so
small that there are no other addresses.

Standalone uses the path of a slapos binary passed as a parameter to
run slapos commands with supervisord or as subprocesses, instead of
recreating an entry point script into the slapos version currently
running standalone. Normally the slapos binary provided is actually
the same slapos. One reason to do it this way is that in test nodes,
standalone is run with pygolang's gpython which doesn't support
running buildout (buildout tries to reexec with -S).

Store ip/netmask instead of ip/prefixlength in computer definition file
generated by standalone because that is what format expects to parse.

Due to typo, ipv4 address was wrongly used when loading ipv6 range
from computer definition file. Also, prefixlength was misleadingly
loaded as netmask.

if ipv6 is a range with prefix length N < 112:
  => give a unique ipv6 to each partition and a unique /N+16 range
if ipv6 is a range with prefix length >= 112
  => give a unique ipv6 to each partition
if ipv6 is not a range:
  => give the same ipv6 to each partition
Co-authored-by: Xavier Thompson <xavier.thompson@nexedi.com>

See merge request nexedi/slapos.core!540

   This is covered by UI test

    This ensure that when users access the site, the information is likely to be cached already.

  Dont fetch the list twice for Network and Organisation
  We dont use Project anymore
  Direct call API for getAccessStatus instead call ComputeNode_getNewsDict

  Similar to getComputeNodeInformation, use caching system to not
  recalculate status for the related software instances, only acquire
  from cache if it is updated.

  There isn't a need for such aggregated monitoring on Project

See merge request !546

    This is mostly covered by UI tests

See merge request !545

See merge request !543

See merge request nexedi/slapos.core!542

  There are really slow servers running tests, if they fail it cause way too many false positives.

  On a distributed tests environment, it is OK, since the fast servers may pick all the tests anyway,
  causing a bad node be ignored.

  Non owners cannot transfer the objects across different projects.

  It is safe to ignore if value is none (it indicates the user sent a raw data).

See merge request !515

See merge request !541

  Priority is set to avoid large accumulation of activities to be queued.

Use the format return code to distinguish between failure to format or
failure to report to master at the end. Then make boot run format even
if the master is unreachable, and:
* if format fails to report, wait until master is reachable, and retry
* if format fails for another reason, retry after a fixed 15s delay
* once format succeeds fully, bang

See merge request nexedi/slapos.core!537