[2025sl93041-gkb]

Summary

Closes #702.

When any formatting function (format_decimal, format_currency,
format_percent, format_compact_decimal, etc.) received an invalid
string value (e.g. "not_a_number"), the internal
decimal.Decimal(str(value)) call would raise a raw
decimal.InvalidOperation — a third-party implementation detail from the
decimal standard library module — rather than Babel's own
NumberFormatError.

This is a leaky abstraction: callers who want to catch bad-value errors
need to import and handle decimal.InvalidOperation in addition to (or
instead of) NumberFormatError.

Root cause

NumberPattern.apply() (and the internal _get_compact_format()
helper) both do:

if not isinstance(value, decimal.Decimal):
    value = decimal.Decimal(str(value))

decimal.Decimal raises decimal.InvalidOperation for strings it
cannot parse. No surrounding try/except was present.

Fix

Wrap the conversion in a try/except block and re-raise as
NumberFormatError:

if not isinstance(value, decimal.Decimal):
    try:
        value = decimal.Decimal(str(value))
    except decimal.InvalidOperation as e:
        raise NumberFormatError(f"{value!r} is not a valid number") from e

Both NumberPattern.apply() and _get_compact_format() are fixed.

Test

Added test_number_pattern_apply_invalid_string_raises_number_format_error
in tests/test_numbers.py. The test calls NumberPattern.apply()
directly (via parse_pattern('0.###')) with a mock locale so it works
without requiring the full CLDR data build.