Fix ParserError on subsecond precision beyond nanoseconds

[afonsojanu]

Summary

pendulum.parse() raises ParserError for an otherwise valid-looking timestamp when the fractional-seconds part has more than 9 digits, e.g.:

>>> pendulum.parse("2001-01-01T12:34:56.1234567890Z")
Traceback (most recent call last):
  ...
pendulum.parsing.exceptions.ParserError: Unable to parse string [2001-01-01T12:34:56.1234567890Z]

datetime.fromisoformat() has no such limit and just truncates to microseconds:

>>> datetime.fromisoformat("2001-01-01T12:34:56.1234567890+00:00")
datetime.datetime(2001, 1, 1, 12, 34, 56, 123456, tzinfo=datetime.timezone.utc)

Reproduces with the "common" datetime format too (this path has no Rust equivalent, so it's hit regardless of the compiled extension):

>>> pendulum.parse("2016/10/06 12:34:56.1234567890")
pendulum.parsing.exceptions.ParserError: Unable to parse string [2016/10/06 12:34:56.1234567890]

Root cause

Both the pure-Python ISO 8601 parser (pendulum/parsing/iso8601.py, used as a fallback when the compiled extension isn't available, e.g. PENDULUM_EXTENSIONS=0) and the "common" format parser (pendulum/parsing/__init__.py, always pure Python) match the subsecond group with \d{1,9}. A 10th digit makes the whole regex fail to match, so the entire string is rejected.

Both call sites already truncate the captured group to 6 characters a few lines below before converting it to microseconds:

subsecond = m.group("subsecond")[:6]
microsecond = int(f"{subsecond:0<6}")

so the {1,9} cap wasn't protecting anything downstream, it was just rejecting valid input before it ever got there.

Fix

Widened both regexes from \d{1,9} to \d+. The existing truncate-to-6-and-pad logic already does the right thing for any number of digits, so no other changes were needed.

Tests

• Added test_rfc_3339_extended_beyond_nanoseconds and test_common_format_extended_beyond_nanoseconds to tests/parsing/test_parsing.py, covering the public parse() API for both the ISO 8601 and "common" format paths.
• Added test_parse_iso8601_subsecond_beyond_nanoseconds_pure_python to tests/parsing/test_parse_iso8601.py, importing pendulum.parsing.iso8601.parse_iso8601 directly so the pure-Python parser is exercised even when the compiled extension is installed (the Rust parser doesn't have this bug since it isn't regex-based).

Confirmed all three new tests fail with ParserError against the unpatched regexes and pass after the fix. Ran the full test suite (pytest tests/, with the Rust extension built via maturin develop --release): 1850 passed, 3 skipped (pre-existing, unrelated to this change), no new failures or warnings.

Closes #935.