fix(pack): byte-counted slash construct must not re-encode as UTF-8

[fglock]

Summary

Fixes a bug in pack's "len/string" slash construct where the length prefix was computed via str.getBytes(UTF-8) while the payload was written via PackWriter.writeString() using ISO-8859-1 (one byte per Latin-1 char). For a byte string of high-Latin-1 bytes (e.g. the output of utf8::encode), this produced a doubled length prefix and zero-padded payload.

For a 12-byte string, the bug produced:

19000000 c3a9...c3a9 00 0000000000000000000000

instead of the correct:

0d000000 c3a9...c3a9 00

Repro before the fix

./jperl -e '
  my $v = "\xc3\xa9" x 6;
  print unpack("H*", pack("V/Z*", $v)), "\n";
'
# 19000000c3a9c3a9c3a9c3a9c3a9c3a900000000000000000000000000   (broken)
# 0d000000c3a9c3a9c3a9c3a9c3a9c3a900                           (perl 5.42)

Why it matters

BSON::PP round-trips every non-ASCII string through pack("V/Z*", $utf8), so this bug broke essentially every BSON corpus and mapping test for any string containing 2- or 3-byte UTF-8. ./jcpan -t BSON goes from 71/72 failing test files to 12/72.

Fix

Match the byte/char count writeString() actually uses: in byte mode use the ISO-8859-1 byte count, otherwise use the Java char count (which matches writeString's per-char output for plain Latin-1 and its per-codepoint output for high-Unicode strings).

Test plan

• Added src/test/resources/unit/pack/slash_string.t covering V/Z*, V/a*, V/A*, n/Z*, round-trip, and the exact BSON wire-format shape (7 subtests, all pass).
• make (full unit test suite) — passes.
• Manually re-ran ./jcpan -t BSON: failing test files dropped from 71 to 12. Remaining failures are unrelated upstream issues (dualvar / B::svref_2object flag bits, Math::BigFloat::as_number('zero') deprecated API on the 32-bit-int fallback path, NaN/Inf encoding edge cases).

Generated with Devin