[GriffinCanCode]

Fixes #158780.

Unstructured headers discard the defects already recorded in their parse tree.
For example, HeaderRegistry()('Subject', '=?utf-8?b?YQ?=') decodes to 'a' but
reports no InvalidBase64PaddingDefect through its public defects attribute.

Propagate those defects and avoid repeated sibling-list concatenation in the
existing collector. Keep defect order, duplicate object identity, and empty-tree
behavior. The existing parameterized tests now check decoding, defects, and
folding for both Subject and custom headers.

This omission also appears in catalog item 34b.
The collector is listed in #136063, alongside the active parser work in #152521
and #152710. Neither PR currently implements these two changes; this is a small
fix in the same area, with that prior work acknowledged.

Validation on macOS arm64, CPython 3.16 debug:

• ./python.exe -m test test_email test_mailbox test_smtplib: 2,305 tests, 5 skips.
• make patchcheck and the applicable pinned pre-commit hooks pass.
• Decoding and folding match in 78 comparisons across three policies. All five
  added runtime lines were profiled; bounded measurements checked CPU, wall
  time, and traced allocations without claiming universal speed improvements.
• The full 561-module run completed with local failures in test_chown_gid,
  three test_glob_dotdot cases, and test_fast_path_avoid_busy_loop. The first
  two reproduce with the original email package; the subprocess assertion
  passes in a focused rerun. The full run is not clean.

AI assistance: Codex helped investigate the behavior, prepare the patch and
tests, and run the checks and profiling.

• Issue: UnstructuredHeader drops encoded-word Base64 padding defects #158780

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[NekoAsakura]

The issue has been closed as not planned.