(Upstream is https://github.com/python/typing)

- Add TYPE_CHECKING (false at runtime, true in type checkers) (upstream python#230).
- Avoid error on Union[xml.etree.cElementTree.Element, str] (upstream python#229).
- Repr of Tuple[()] should be 'Tuple[()]' (upstream python#231).
- Add NewType() (upstream python#189).

…hon#189

Closes the spec-extrapolation gap surfaced by pythia python#189 2026-04-27:
kwnames_ probe (c2f255e, 1 LOC net) demonstrates single-commit
class-state migration WORKS for low-site-count fields. It does NOT
extrapolate to high-site-count fields (code_ at 46 sites, preloader_
at 28, current_func_ at ~30). 'The first thread pulled proves only
that thread can be pulled.'

Changes:

§4.A.5 step 3 reframed: same-commit-deletion = no parallel-state
interval = no equality DCHECK needed. ONLY for ≤5-site-class fields.

§4.A.5b (NEW): explicit DO-NOT-EXTRAPOLATE acknowledgment + per-field
complexity bracket (≤5/6-15/>15 sites → single/medium/multi-commit
with JIT_DCHECK equality required for the latter two) + pre-Pilot-5
per-field site-count audit grep + failure-mode predictions per
bracket.

§4.C Pilot 5: per-field bracket with measured site counts, mandatory
pre-audit step, JIT_DCHECK equality REQUIRED for all 4 remaining
fields (each is >5-site-class). Estimate revised from 25-40 commits
uniform to 13-19 commits per measured bracket.

Authored 2026-04-27 per pythia python#189 + supervisor 2026-04-27T10:13:10Z
(a)+(b)+(c)+(d) framing + theologian §4.C per-field bracket addition.