grep: detect format flags bundled inside short-flag clusters

[EtaCassiopeia]

Follow-up to #2543 — this fixes the second part of it.

has_format_flag checks extra_args for things like -l/-c/-o so it knows to skip the usual match-line parsing and just pass rg's output straight through (since those flags change rg's output format entirely). That check only matches exact tokens though, and extract_pattern_path merges plain boolean short flags into one combined token before they get here — so -rln becomes -ln after r is stripped, and -ln never matches -l exactly.

End result: something like rtk grep -rln pattern path runs rg with -l correctly, but then tries to parse the (now filenames-only) output as if it were normal file:line:content matches, finds nothing that fits that shape, and reports "0 matches" even when there clearly are matches. That's worse than just erroring, since it looks identical to a real empty result.

Fix is small — has_format_flag now also scans the individual letters of a short cluster (long flags and exact single-flag matches are unaffected). Added a few tests for the clustered case plus one to make sure long flags like --ignore-case don't get walked character-by-character by mistake.

Tested locally:

• rtk grep -rln pattern path now correctly falls through to passthrough and shows the real file list instead of "0 matches in 0 files"
• rtk grep -rn pattern path unaffected
• existing grep_cmd test suite passes (72 tests)

closes #2543

[CLAassistant]

All committers have signed the CLA.