more smp fixes

[wjhun]

• remove sloppy uses of ci->have_kernel_lock on lock acquisition; flag should instead be used for debug assertions
• explicitly release kernel lock on calls to runloop (aside from kernel page fault)
• clear ci->have_kernel_lock before suspending kernel context on page fault, reset on context resume
• allow pagecache I/O completions to be served on either bhqueue or runqueue depending on origin of request
• change vq_interrupt to enqueue to vq->sched_queue instead of bhqueue
• fix invalid dereference of current in syscall_schedule

While stability issues remain when running with SMP (see #1302 for known issues), webg and other tests progress much further now.

[francescolavra]

Looks good to me.