fix(instances): a new VM's CPU is a mode or a model the host offers, checked by libvirt

- A named CPU model is written without check='none' and with
  fallback='forbid': libvirt checks before the start that the host runs
  it, and it is not swapped for another model.
- Mode maximum is written as that mode; it was written as a model named
  maximum.
- Model names up to 64 characters: the -noTSX and -IBRS models, the ones a
  mixed set of hosts needs, were refused at 20.
- The mode or model is checked against the host's capabilities when the VM
  is defined, from the page and the API alike; no CPU element (QEMU's
  oldest default CPU) is no longer offered.
- The default CPU mode setting offers the modes the new VM dialog has;
  no-model and custom, which matched none of them, become host-model.

instances/models.py

@@ -352,7 +352,8 @@ class CreateInstance(models.Model):
     vcpu = models.IntegerField(
         error_messages={"required": _("No VCPU has been entered")}, validators=[MinValueValidator(1)]
     )
-    vcpu_mode = models.CharField(max_length=20, blank=True)
+    # a mode of libvirt's (host-model, host-passthrough, maximum) or a CPU model's name
+    vcpu_mode = models.CharField(max_length=64)
     disk = models.IntegerField(blank=True)
     memory = models.IntegerField(
         error_messages={"required": _("No RAM size has been entered")}, validators=[MinValueValidator(1)]