gh-155504: Set the interpreter before fallible thread state initialization - #155590

alloc_threadstate() returned a reset thread state with base.interp still
set to NULL. If one of the later PyStats, TLBC, or QSBR initialization steps
failed, new_threadstate() passed that partially initialized state to
free_threadstate(). The cleanup path needs the interpreter both to identify
the embedded initial thread state and to release interpreter-owned resources.

Set base.interp in alloc_threadstate() before any later initialization can
fail, so every error path can safely free or recycle the thread state.

The change also adds a Core and Builtins NEWS entry.

Fixes #155504.

Tests

• Py_DEBUG build:
  ./python.exe -m test -j4 --timeout=180 test_capi test_interpreters test_threading test_embed
  (2,001 tests passed)
• Py_DEBUG free-threaded build with PyStats:
  ./python.exe -m test -j4 --timeout=180 test_pystats test_capi test_interpreters test_threading test_embed
  (1,836 tests passed; test_interpreters was skipped because the GIL was disabled)
• Temporary failure injection verified cleanup for both the embedded/preallocated
  and heap-allocated thread state paths; the injection was removed afterward.
• git diff --check

• Issue: Logic bug in new_threadstate() error path, interp is NULL #155504