Audit I/O code for failure-safety

[brson]

Code that interacts with the scheduler is wildly unsafe and probably not being properly defensive about recovering from failure or at least rtaborting on unexpected failure.

In particular, in the current iteration, I/O blocking is subject to linked task failure, under which blocked tasks my resume and fail, and this is not recovered correctly. In the short term we should make all I/O blocking non-interruptable, then handle interruptability on a case-by-case basis.