xen/intr: move handler removal to isrc release

[ehem_freebsd_m5p.com]

Summary

xen_intr_release_isrc() should be an internal function cleaning
*everything* up. As such having a handler should not be a problem and
what could actually be concerning is a non-zero reference count.

[ehem_freebsd_m5p.com]

This seems rather more appropriate. xen_intr_release_isrc() now releases everything and internal functions no longer need to call xen_intr_unbind() to get the handler removed. One could add a KASSERT() for isrc->xi_refcount == 0, this would require xen_intr_release_isrc() to zero the count in case of failure though.

[ehem_freebsd_m5p.com]

Odd thing is, as I look at this more I wonder whether this is needed at all. Real problem was how xen_intr_bind_virq() and xen_intr_alloc_and_bind_ipi() were handling processor assignment. Merge everything into xen_intr_bind_isrc() and this may no longer be needed.

[ehem_freebsd_m5p.com]

This is looking odd enough to perhaps not actually get in...

[ehem_freebsd_m5p.com]

Things got rearranged in a hopefully better way, so marking this as dead. Pretty sure underlying issues got fixed.