Feature: infer output format from --output file extension

[vmvarela]

Context

--output result.json still writes CSV unless -O json is also passed. Input files already auto-detect format by extension (data.parquet, data.yaml — no -I needed); output should behave symmetrically.

Proposed behavior

sql-pipe data.csv -O json --output result.json 'SELECT * FROM t'   # today: both flags required
sql-pipe data.csv --output result.json 'SELECT * FROM t'           # proposed: .json implies -O json

• When --output <file> is set and -O/--output-format is not, infer the format from the extension: .csv .tsv .json .ndjson .xml .md .html .sql.
• Explicit -O always wins over the extension.
• Unknown or missing extension → current default (csv), no error.

Excluded (YAGNI)

• Inferring format for stdout (no filename — keep csv default)
• Warning when extension and -O disagree (explicit flag wins, silently)

Acceptance criteria

• --output r.json writes JSON without -O
• -O tsv --output r.json writes TSV (flag wins)
• --output r.txt falls back to csv
• Same mapping table reused from input extension detection if practical
• Help text for --output mentions the inference
• Integration test