Change os_unfair_lock calls to use a dispatch queue instead.

[kstenerud]

Goal

#832 shows a crash in os_unfair_lock_unlock that couldn't be traced back to anything user-side. Since we can't repro the issue, we need another way to defend against this.

Design

os_unfair_lock is a relatively new API, so it's not inconceivable that there could still be bugs in it. Dispatch queues, on the other hand, have been battle-hardened over decades. Dispatching to a local queue on a local thread has very low overhead, and will introduce far less latency than calling NSProcessInfo, so this will on the whole run faster.

Changeset

bsg_mach_headers_add_image now uses a dispatch queue rather than os_unfair_lock or OSSpinLock.

Testing

Re-ran all unit tests, and did manual tests to ensure end-to-end symbolication.

[nickdowell]

Looks good to me