Check for errors from str_path_to_cfstring_ref.

[jorendorff]

Closes #301.

[0xpr03]

Are you sure this fixes it or rather just works in case the race condition for our runtime panic isn't triggered ?

[0xpr03]

we should probably investigate whether this should be handled instead of ignored (and deallocated)

[jorendorff]

Are you sure this fixes it or rather just works in case the race condition for our runtime panic isn't triggered ?

Medium-high confidence that this fixes the actual problem. Here's the evidence:

• Before the change, the test crashed pretty reliably for me, say 95-98% of test runs, except in lldb. The first thing I tried is just adding assert!(!cf_path.is_null()) there, and it stopped the crashes immediately; they became regular Rust panics.

• While I was reducing this from my real application, I ended up making 8 versions of the test (each containing fewer and fewer random bits of my app) and the change, backported to 4.0, makes them all pass.

To get a bit more confidence, one could write a program that passes a null second argument to cf::CFArrayAppendValue, to see if it crashes with this strange message about a "foreign exception".