OPTIM: poll: optimize fd management functions for low register count CPUs

OPTIM: poll: optimize fd management functions for low register count CPUs

Looking at the assembly code that updt_fd() and alloc/release_spec_entry
produce in the polling loops, it's clear that gcc has to recompute pointers
several times in a row because of limited spare registers. By better
grouping adjacent structure updates, we improve the code size by around
60 bytes in the fast path on x86.