feat: positional file arguments and multi-file joins

[vmvarela]

Summary

Closes #155.

Accept files as positional arguments. Each file becomes a table named after its basename (sans extension). Stdin remains table t.

Examples

# Single file — no more cat
sql-pipe orders.csv 'SELECT * FROM orders WHERE amount > 100'

# Multi-file join
sql-pipe orders.csv customers.csv \
  'SELECT c.name, SUM(o.amount) FROM orders o JOIN customers c ON o.cust_id = c.id GROUP BY c.name'

# Stdin still works, still called t
cat data.csv | sql-pipe 'SELECT * FROM t'

# Mix stdin + files
cat events.csv | sql-pipe users.csv 'SELECT * FROM t JOIN users ON t.uid = users.id'

Acceptance Criteria

• Files passed as positional arguments are loaded as tables
• Table names are derived from filenames (basename sans extension)
• Stdin input is still available as table t
• Multi-file joins work correctly
• Input format is auto-detected from file extension (.csv, .tsv, .json, .ndjson, .xml)
• File-vs-query disambiguation works (heuristic: if arg is a readable file, treat as input; query is last non-file argument)
• -- separator works for disambiguation
• All existing tests pass
• New tests cover single-file, multi-file, stdin+file, and edge cases (155a–155i)

Implementation Notes

• format.zig: Added InputFormat.fromExtension() for auto-detection from file extension
• sqlite.zig: Parameterized createTable/prepareInsertStmt/createAllTextTable with table_name
• args.zig: Added FileInput struct, file-vs-query disambiguation, -- separator support
• loader.zig: Generalized loadCsvInput to accept table_name and file_path
• json.zig/xml.zig: Parameterized loaders with table_name
• main.zig: Orchestrates multi-file loading with poll()-based stdin detection
• Documentation updated in README.md and man page