bus_dmamap_create() leaks the freshly kmalloc'd map when the initial bounce-page allocation partially fails (vkernel never got pc64's kfree(*mapp) error-path fix)

Summary

*mapp kmalloc'd before the bounce top-up; ENOMEM returns with map_count not incremented and the map never freed - callers treat failed create as no-map, leaking 72B M_DEVBUF per call, exhaustible to vkernel DoS under lowaddr pressure + NOWAIT creates. pc64 fixed exactly this; the fork predates it. kfree-and-NULL-on-error fix.