str: fix the character count of a stepped slice, which reversed() read past - #8525

A stepped slice of a non-ASCII string reports a character count one too high whenever the
span is an exact multiple of the step, and the overstated count makes reversed() on the
result read past the end of the buffer and panic:

s = "".join(["a", "é", "c"])
list(reversed(s[::3]))
# thread 'main' panicked at crates/common/src/str.rs:327:37:
# index out of bounds: the len is 1 but the index is 1

The subject goes through a variable because a constant subscript is folded at compile time,
and the folded path is correct -- only a slice evaluated at runtime reaches the bug.

Cause

do_stepped_slice and do_stepped_slice_reverse compute the result's character count as
(range.len() / step) + 1. That is the number of steps taken plus one, which is right
only when the last step lands short of the end; when the span divides evenly it counts a
character the slice never collected.

ASCII subjects are unaffected: that arm collects into an AsciiString, whose length comes
from the data rather than from the formula.

The count is passed to new_with_char_len, whose contract is that it is accurate, and is
stored as the string's character length. So the result is a string that claims a character
its buffer does not hold -- len() lies, and any consumer that walks to the last index
walks off the end. reversed() is the one that panics, because it indexes from
char_len - 1.

Fix

Use range.len().div_ceil(step), which is the number of elements the underlying range
yields.

Verification

• New assertions in extra_tests/snippets/builtin_str_unicode_slice.py, covering both step
  directions over 2-, 3- and 4-byte characters, and reversed() on each result. They fail
  on main at the first case (AssertionError: ('aéc', 3, 2)) and pass here, so they are
  not vacuous -- the subject goes through a function to keep the compiler from folding the
  subscript.
• A differential over 44331 subscript and slice shapes -- every combination of start,
  stop and step from {None, 0..6, -1..-6, ±100} over strings straddling the ASCII split,
  the WTF-8 surrogate range, the astral plane and the 64-code-point index-group boundary --
  matches CPython 3.14.6 exactly. On main the same run differs on 3100 lines, all of
  them stepped slices, all of them a wrong length beside correct content.
• test_str test_string test_re test_bytes test_json: 893 run, 44 skipped, SUCCESS.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Corrected character counts for stepped Unicode string slices when the step does not evenly divide the selected range.
  • Improved slicing reliability for accented, Korean, Hebrew, and emoji text.
  • Fixed reverse stepped slices to report accurate lengths and avoid incorrect results.

• Tests

  • Added regression coverage for positive and negative steps, including slice content, counts, and reverse iteration safety.