…126)

DNF5 (Fedora 41+, RHEL 10) removed the --add-repo argument from
config-manager. Use curl to drop the .repo file directly into
/etc/yum.repos.d/, which works on both DNF4 and DNF5.

- Update intro: 'reads CSV from stdin' → 'reads CSV, JSON, or NDJSON'
- Add JSON/NDJSON input paragraph and example to Usage section
- Add -I/--input-format and -O/--output-format to flags table
- Fix broken chain example (missing -H on first sql-pipe call)
- Add Real-world examples section with 6 verified live examples:
  La Liga wins/goals (CSV), OWID energy solar/wind (CSV+chain),
  restcountries density (NDJSON), Open-Meteo Madrid forecast (NDJSON)

Closes #127

* feat: add --silent / -s flag to suppress row count output (#114)

* docs: add --silent / -s flag to README and man page

…ry (#88)

* feat: add --validate mode to check CSV syntax without running a query (#88)

* feat: extend --validate to support JSON and NDJSON input formats

* fix: enforce --validate mutual exclusion with --output and --columns

* feat: add --sample <n> flag for quick data preview with schema (#89)

* ci: downgrade labeler to v5 to restore title-based label matching

* ci: replace broken labeler action with github-script title matching

* fix: address code review findings in --sample implementation

- Add SampleWithOutput error: --sample --output now exits 1 with error
  instead of silently discarding the output file path
- Fix stdout write errors in runSample to call fatal instead of
  std.log.err, preventing silent truncation with exit code 0
- Add type_inference field to SampleArgs so --no-type-inference is
  respected in --sample mode (shows all columns as TEXT)
- Use @max for max_col_width computation (style)
- Clarify help text ambiguity around mutual exclusions
- Add 10 integration tests (85-94) covering all --sample acceptance
  criteria: row count, schema on stderr, type inference, TSV input,
  error cases and edge cases

* feat: support multi-character input delimiters (#86)

Extend -d/--delimiter to accept strings of 1-8 bytes instead of a
single character. Common real-world separators like '||', ';;', or
two spaces now work without preprocessing.

- CsvReader: delimiter field changed from u8 to []const u8; added
  partial_delim: usize to track in-progress multi-byte matches in
  the streaming state machine
- parseDelimiter: returns []const u8, rejects empty and >8-byte values
- writeField / printRow / printHeaderRow: delimiter type updated to
  []const u8; quoting detection uses std.mem.indexOf instead of
  byte-by-byte comparison
- New unit tests: 2-char (||), 3-char (;;;), partial-match false
  positive, quoted field containing multi-char delimiter
- New integration tests 95-98: double-pipe, three-char, empty-error,
  too-long-error
- README and man page updated to describe the 1-8 char constraint

* test: add edge-case unit tests for multi-char delimiters

- Reset partial_delim at start of nextRecord (latent correctness fix:
  avoids stale state if a previous call exited via a non-fatal error)
- Add 6 unit tests covering: empty first field, empty last field,
  only-delimiter input, EOF without newline, partial delimiter at EOF
  treated as field content, and greedy left-to-right matching behavior

Closes #99

- Row-based XML parser and writer (no external library)
- `-I xml` / `-O xml` format flags
- `--xml-root` / `--xml-row` element name customisation with XML name validation
- XML supported in `--columns` and `--validate`; rejected with clear error in `--sample`
- `src/sqlite.zig`: shared SQLite helpers extracted from json.zig (DRY)
- 13 unit tests + 15 integration tests (entities, NULL, roundtrip, edge cases)

On Windows there is no system sqlite3 library, so xml.zig unit tests
fail with 'unable to find dynamic system library sqlite3'.
The SQLite amalgamation is already downloaded before this step,
so bundling it is both correct and sufficient on all platforms.

Closes #136