Add --select, --rename, and --cast flags for load-time column transformation

[vmvarela]

Description

Add flags to select, rename, and cast columns during loading, before data reaches SQLite.

# Select specific columns only
sql-pipe data.csv --select 'name,amount' 'SELECT * FROM t'

# Rename columns during load
sql-pipe data.csv --rename 'old_name:new_name,amount:revenue' 'SELECT new_name, revenue FROM t'

# Cast column types (override inference)
sql-pipe data.csv --cast 'zip:TEXT,amount:REAL' 'SELECT * FROM t'

# Combine all three
sql-pipe data.csv --select 'name,zip,amount' --rename 'amount:revenue' --cast 'zip:TEXT' 'SELECT * FROM t'

Motivation

Type inference is imperfect — ZIP codes like "01234" get detected as INTEGER (losing the leading zero). Column renaming is needed when source data has spaces, special characters, or non-SQL-friendly names. Currently users must use CAST() in every SQL query, which is verbose and must be repeated.

Acceptance Criteria

• --select <cols> flag: load only specified columns (comma-separated)
• --rename <mapping> flag: rename columns during load (old:new pairs, comma-separated)
• --cast <mapping> flag: override inferred types (col:TYPE pairs, comma-separated)
• Flags can be combined in a single invocation
• --select preserves column order as specified
• --cast supports: INTEGER, REAL, TEXT, DATE
• Error on unknown column names in --rename and --cast
• Integration tests for each flag and combinations
• Help text updated

Implementation Notes

• Intervene between header parsing and table creation in the loader pipeline
• --select: filter column names before type inference and INSERT binding
• --rename: apply name mapping after header parsing
• --cast: override type inference results before CREATE TABLE
• The PRAGMA table_info approach (already in getTableColumns) provides column list for validation
• Multi-character delimiters make index-based column skipping in the CSV parser more complex — consider post-filtering the record array instead