kern: factor out the 'writing' bits of user process coredumping

[kevans]

Summary

Instead of assuming we have a vnode to dump to, pull out the bits that
describe how to write acoredump into a struct coredump_writer; the ctx in
that is expected to be opaque, but used by write/extend functions. This
should not be a functional change- we change two callers to use the classic
vnode write/extend interface.

This opens us up to other possible targets for coredumps, such as a shmfd.

[kib]

sys/kern/kern_exec.c

2002

Move these functions to hypothetical vnode dumper source file.

[kevans]

sys/kern/kern_exec.c

2002

I left these alone here and moved them along with everything else in the next commit, since that diff had ended up a bit larger than I expected.

[kevans]

Add an init_fn to the coredump_writer

This gives the caller a chance to do some lazy initialization or record the
coredump parameters. It will be documented in coredumper_register(9) in a later
change in this stack.

The vnode dumper could probably move vnode setup/corefile_open to an init
function and avoid leaving a 0-byte .core file around if sv_imgact is just going
to fail in the early stages anyways (racct/limit checking, compressor_init), but
I decided that that's probably more of a cosmetic issue and punted for now.

[kevans]

ninja edit: missed a hunk in patch transfer, s/return/goto done/ if init_fn fails