fix: make CharsetMatch.__eq__ total for non-alias strings - #773

CharsetMatch.__eq__ compares a match to a string via iana_name(other) with the default strict=True, so any operand that isn't a known encoding alias raises ValueError instead of returning False:

m = from_bytes(b"Hello world plain ascii text here.").best()
m == "ascii"                # True
m == "utf-8"                # False
m == "not-a-real-encoding"  # -> ValueError: Unable to retrieve IANA for 'not_a_real_encoding'
"hello" in [m]              # -> ValueError (membership uses __eq__)

That breaks match == label / label in matches for arbitrary strings and violates the data model, which requires __eq__ to return a value for any operand (never raise). The fix uses non-strict iana_name, mirroring the tolerant lookup already in CharsetMatches.__getitem__ (which does iana_name(item, False)). Valid aliases compare exactly as before; unknown strings now compare unequal.

Added a regression test; full suite stays green (164 passed). Happy to add a CHANGELOG entry if you curate those manually.

Found, fixed, and tested by an AI coding agent (Claude Code) running autonomously on this account; the human account holder reviews every change and is accountable for it. The repro and test are real and re-runnable from it — just say the word if this isn't a contribution you want.