#1618 fix: UTF-16BE skill files keep their byte order during setup

Summary

• UTF-16 big-endian skill files now keep their byte order when valid content resembles a little-endian carriage return.
• BOM-less UTF-16 line-ending detection remains unchanged.

User Impact

• Previously, a UTF-16BE text skill file with an explicit BOM could be routed through the little-endian heuristic when its payload contained bytes such as 0D 00. Setup could then rewrite valid content using the wrong byte order.
• Explicit UTF-16 BOMs are now authoritative, so generated skill copies preserve their declared encoding while still normalizing CRLF to LF.

Root Cause

• The little-endian BOM and line-ending heuristic were combined in the first branch. A BE BOM did not match the LE BOM check, but the LE heuristic still scanned the BE payload before the explicit BE BOM branch ran.

Changes

• Evaluate explicit little-endian and big-endian BOMs before either BOM-less line-ending heuristic.
• Keep the existing LE-then-BE heuristic order for content without a BOM.
• Add a regression test containing UTF-16BE U+0D00 followed by CRLF.
• Add BOM-less LE and BE characterization tests to protect the existing heuristic paths.

Verification

• TDD Red: the new ambiguous UTF-16BE case failed while the other 8 normalizer tests passed.
• TDD Green: SkillFileContentNormalizerTests 9 passed.
• Repo-local Unity compile: 0 errors, 0 warnings.
• SkillInstallLayoutTests: 18 passed.
• ToolSkillSynchronizerTests: 52 passed.
• git diff --check.

Compatibility

• Only local skill file byte normalization changes. Wire formats and protocol declarations are unchanged.