fix: avoid mutating global csv dialect in run_csv_view fallback

[zza-830]

Problem

When csv.Sniffer().sniff() fails in run_csv_view(), the fallback code calls csv.get_dialect('excel') which returns the class itself (not an instance), then mutates .delimiter on it. This permanently corrupts the global excel dialect for the process lifetime, affecting all subsequent CSV operations.

Closes #47

Fix

Use the built-in csv.excel_tab and csv.excel dialects directly based on the detected delimiter, avoiding any mutation of global state.

# Before (mutates global dialect)
dialect = csv.get_dialect("excel")
dialect.delimiter = delimiter

# After (uses built-in dialects, no mutation)
dialect = csv.excel_tab if delimiter == "\t" else csv.excel

Testing

tests/test_data.py::test_run_csv_view_file_not_found PASSED
tests/test_data.py::test_run_csv_view_empty_file PASSED
tests/test_data.py::test_run_csv_view_shows_limited_rows PASSED