Fix DimensionToken serialization for scientific-notation-like units

[gaoflow]

Problem

DimensionToken._serialize_to escapes a unit's leading e/E so the unit does
not merge with the number into a scientific-notation number token when the
output is re-parsed. The guard is both incomplete and case-corrupting, so
serialize(parse(css)) == css is violated for some valid dimension tokens:

>>> from tinycss2 import parse_component_value_list as p, serialize
>>> serialize(p(r'1\45 '))     # value 1, unit 'E'
'1\\65 '                       # re-parses to unit 'e' -> case flipped
>>> serialize(p(r'1\65 3'))    # value 1, unit 'e3'
'1e3'                          # re-parses as the NUMBER 1000.0

Two failure modes:

• The escape is hard-coded as \65 (lowercase e), so a unit E round-trips
  to e (case flip).
• The trigger only covers e/E and e-/E-, missing e/E followed by a
  digit (e3, E3, e05, ...). Those units merge into the exponent and the
  dimension silently becomes a number with a different value.

Common units (em, ex, EX, Q, ...) are unaffected.

Fix

Escape the leading letter using its own code point (\\%X ) so its case is
preserved, and widen the trigger to any unit whose first character is e/E
followed by end-of-unit, a sign, or a digit.

Tests

test_serialize_dimension_scientific_notation round-trips the problem units
(e, E, e3, E3, e05, e-3, E-3), asserting they stay dimensions with
the same value/int_value/unit, and checks that common units are not over-escaped.

Full suite: 16976 passed. ruff check clean.

[liZe]

Hi!

Thanks for the pull request.

Out of curiosity: did this problem happen in one of your projects?