[dajiaohuang]

Problem

ptags --unsorted passes --sort=no to the workers, but not to the separate ctags invocation that creates the tag-file header. A source file defining zeta before alpha therefore produces those tags in that order while advertising !_TAG_FILE_SORTED\t1. Consumers use this field to choose sorted-file lookup, so the header must describe the actual output.

Change

Pass the same unsorted setting when generating the header. Keep the default sorted behavior unchanged. Add public run_opt integration coverage using a temporary C source and explicit input list/output/directory, checking both the metadata and tag order.

Validation

• On the unchanged base, the new unsorted regression fails; default sorted coverage passes.
• With this change, both tests pass with real Universal Ctags 5.9.0 on Linux and 6.1.0 on Windows: cargo test --locked --test tag_sort -- --test-threads=1.
• cargo test --locked --no-run, changed-file rustfmt check and git diff --check pass.
• Full existing runtime tests were not run: they enumerate the checkout and require its repository/LFS/submodule fixtures. Separate synthetic CLI acceptance covered tracked/untracked/ignored selection, exclusions, list/stdin, submodules, LFS, output modes and configuration; this is not a claim that the full upstream suite passed.