Fix regex filter parsing when pattern contains $ end-anchor

[0b0d3]

Summary

Fixes #257

• Regex filters like /pattern$/ use $ as a regex end-of-string anchor, but the parser always treated the rightmost $ as the options separator
• This caused filters like /\.json$/ to fail with UnrecognisedOption, and filters like /^https?:\/\/[a-z]{8,15}\.top\/[a-z]{4,}\.json$/$xhr,3p,match-case to lose the $ anchor from their pattern
• The fix tries each $ from right to left, but only falls back to the next candidate when the pattern to the left looks like a regex (delimited by /). Non-regex filters still propagate option parse errors immediately

Changes

• src/filters/abstract_network.rs: Replace single find_char_reverse with a loop that tries $ candidates right-to-left, with regex-awareness to avoid false fallbacks
• tests/unit/filters/network.rs: Added 3 test cases covering regex with $ anchor + options, regex with $ anchor alone, and regex without $ anchor
• tests/unit/blocker.rs: Updated easylist filter count (4 more filters now correctly parsed)
• tests/unit/engine.rs: Updated serialization hashes to reflect the new parsing behavior

Test plan

• All 262 existing tests pass (including with css-validation feature)
• New test: regex filter with $ end-anchor and options parses correctly
• New test: regex filter with $ end-anchor and no options parses correctly
• New test: regex filter without $ anchor and options still works
• Non-regex filters with invalid options still return proper errors
• Real-world filter lists (easylist, easyprivacy, ublock-unbreak) parse with correct counts

🤖 Generated with Claude Code

[ImpulseB23]

approach looks good, this is basically what i described in my comment above. logic is clean and the tests cover the key cases.

one small thing: the pattern_before.ends_with("$/") branch in the looks_like_regex check might be unreachable. when you're at a given $, pattern_before is everything to its left, so it can't end with $/ unless there's a consecutive $$ in the pattern. not a bug, just potentially dead code. might be worth dropping it or adding a test that hits it if you want to keep it.

otherwise lgtm