libnvme: fix crash on a malformed exclusion entry

[martin-belanger]

Any exclusion entry with an unknown key, a token without =, or an empty value crashes every caller of libnvmf_exclusion_match(): nvme connect-all, nvme-discoverd, and anything else that consults the list.

printf '[exclusions]\nexclusion = bogus=x\n' > /etc/nvme/exclusions.conf
nvme connect-all ...     # SIGSEGV

entry_matches() parsed each entry with a NULL context to keep the match path quiet, but tid_parse() logs those three cases and libnvme_msg() dereferences the context.

The fix threads the caller's real context down instead, and reports a bad entry at DEBUG. Matching runs on every connect attempt, so one static typo must not produce a log line per connect. libnvmf_tid_parse() and libnvmf_tid_parse_strict() now return -EINVAL for a NULL context rather than documenting it as "parse silently", which was never true.

The second commit makes the typo visible where an admin would look for it. nvme exclusion list showed only drop-in names, so the entries of the main hand-edited /etc/nvme/exclusions.conf could not be listed at all. It now prints every list with its entries and marks the ones that can never match:

$ nvme exclusion list
(default)
  nqn=nqn.2014-08.org.nvmexpress:uuid:abc
  bogus=x  # invalid: never matches

user
  transport=tcp;traddr=10.0.0.1

--verbose logs the reason for each one.

Regression tests cover the four malformed forms plus a hostname in traddr, which was silently rejected with no diagnostic before.

[igaw]

Thanks!