Fix block list parser mis-parsing ABP cosmetic filters as domains

[SeanSith]

Summary

BlockListZoneManager.ReadListFile recognizes #/! comments and
||domain^ / @@||domain^ Adblock network-filter syntax explicitly.
Any other line falls through to a "hosts file format" guess: split the
line on whitespace, and if there's a second word, treat that as the
domain. The second word is validated with DnsClient.IsDomainNameValid
before being enqueued, but the first word is never checked at all.

Adblock Plus cosmetic/procedural filters (##, #@#, #?#) attach a
domain list to a CSS/procedural selector that can contain natural
language, e.g. from easylist.to's fanboy-annoyance.txt:

tiktok.com##[aria-label="Scroll to the top"]
daily.dev#?#.bottom-0.fixed:has-text(This site uses cookies)

Neither line starts with # (the # is mid-line), ||, or @@||, so
they hit the hosts-format fallback. Splitting on whitespace picks out
the English word following the domain as the "second word" — to
and site respectively. A bare single-label name is syntactically
valid per IsDomainNameValid, so both get enqueued as blocked domains.
Once in the block zone, blocking is hierarchical, so this silently
blackholes the entire .to and .site TLDs — including, amusingly,
easylist.to itself.

I found this diagnosing a real installation where easylist.to (the
list's own host) and an unrelated .site-hosted domain were both
returning NXDOMAIN, traced via the block zone's EDNS debug extension
(source=block-list-zone; blockListUrl=...; domain=to) back to this
list.

Fix

Require the first word to already look like a domain name or IP
address before trusting the hosts-format split at all:

if (!DnsClient.IsDomainNameValid(firstWord) && !IPAddress.TryParse(firstWord, out _))
    continue;

This isn't specific to ##/#@#/#?# — it closes the same hole for
any current or future Adblock/uBlock/AdGuard syntax that isn't
explicitly handled (scriptlet/JS injection #%#, AdGuard's $$ HTML
filtering, etc.), since all of them put a character outside the
[a-zA-Z0-9-_/.] domain charset somewhere in that first token. No
marker allowlist to maintain.

Testing

There's no test project in this repo, so I verified by re-implementing
the exact parsing logic (comment/||/@@||/hosts-format branches,
IsDomainNameValid's character rules pulled from
TechnitiumLibrary.Net, IPAddress.TryParse via socket.inet_pton)
and running it against 31 block lists from a real deployment's
configured set — before/after enqueued-domain counts:

List	Before	After	Removed
easylist.to/easylist/fanboy-annoyance.txt	2499	2206	293 (incl. to, site)
filters.adtidy.org/.../15.txt	178145	178126	19
29 other configured lists (StevenBlack/hosts, AdAway, OISD, Hagezi, etc.)	—	—	0

Zero regressions on every genuine hosts-format list; the fix only
strips entries from the two lists that use Adblock cosmetic-filter
syntax, and confirmed to/site no longer appear in the parsed output
of fanboy-annoyance.txt after the change.

[ShreyasZare]

Thanks for the PR. Will check it soon.