feat: improve SQL error messages with column context and did-you-mean hint

[vmvarela]

Summary

Closes #81

• On SQL query error (exit 3), print the list of available columns from PRAGMA table_info(t) and suggest the closest match when the missing identifier has edit distance ≤ 2 from an existing column name.
• Fix --json incorrectly rejected when combined with --delimiter / --tsv (those flags only affect input parsing, not output format).
• Fix double-free bug in csv.zig triggered by error.UnterminatedQuotedField when manual cleanup clashed with errdefer blocks.

New behaviour

$ printf 'id,amount,region\n1,100,east\n' | sql-pipe 'SELECT amout FROM t'
error: no such column: amout
  table "t" has columns: id, amount, region
  hint: did you mean "amount"?

$ printf 'id,amount,region\n1,100,east\n' | sql-pipe 'SELECT revenue FROM t'
error: no such column: revenue
  table "t" has columns: id, amount, region

$ printf 'name,age\n"unterminated' | sql-pipe 'SELECT * FROM t'
error: row 2: unterminated quoted field

Changes

• src/main.zig — levenshteinDistance, getTableColumns, printSqlErrorContext, fatalSqlWithContext; all post-createTable SQL errors now use fatalSqlWithContext; --json incompatibility narrowed to --header only
• src/csv.zig — remove manual cleanup before return error.UnterminatedQuotedField (was double-freeing memory already owned by errdefer)
• build.zig — integration tests 26 (column list on SQL error), 27 (did-you-mean hint), 28 (CSV row number), 17b (--json + --delimiter now works)
• README.md — Exit Codes section updated with example output