ruff filter: `is_format` matches a bare `format` token anywhere in argv, so the passthrough cap is applied inconsistently

ruff_cmd.rs:

let is_format = args.iter().any(|a| a == "format");

is_check is decided from args.first(), is_format from any(). So any invocation carrying a bare format token anywhere selects filter_ruff_format for its output — and filter_ruff_format's fallback branch is a bare result.push_str(output.trim()), which skips the truncate(…, passthrough_max_chars) cap the else branch applies.

Reproduction (ruff 0.16.6, LC_ALL=C)

$ ruff help format | wc -c        # 4889
$ rtk ruff help format | wc -c    # 4889   <- uncapped
$ ruff help check | wc -c         # 6297
$ rtk ruff help check | wc -c     # 2001   <- capped at passthrough_max_chars

Two sibling invocations of the same subcommand get different caps, decided by whether the word format appears in argv. Output is correct in both cases — this is over-delivery, not loss — but the routing is accidental.

Suggested direction

Derive it from the same decision is_check_invocation makes:

let is_format = args.first().is_some_and(|a| a == "format");

Predates #3083 (which only changed is_check); it became observable there because non-check subcommands now actually reach the filter.

Start in ruff_cmd.rs by comparing the is_format decision with is_check_invocation, then run the reported ruff help format and ruff help check reproductions. Done means format is selected from the first argument consistently, with passthrough output subject to the intended cap.