Deprecation cleanup, decorator-dependency removal, and startup-time fixes - #15310

…s branch

- Rewrite catch_format_error (formatters), only_when_enabled and
  catch_corrupt_db (history) as plain functools.wraps closures; remove
  the 'decorator' runtime dependency and the types-decorator stubs from
  the extras and the mypy CI install. The two history.py decorators are
  typed as signature-preserving with ParamSpec, matching what the
  types-decorator stubs gave them, so disallow_untyped_decorators stays
  clean and every decorated HistoryAccessor method keeps its inferred
  type.
- completer.py: import TypedDict/Protocol/NotRequired/TypeAlias/
  TypeGuard from typing unconditionally - all are in the 3.11 stdlib,
  which is the minimum supported version. typing_extensions remains
  only for TypeAliasType in guarded_eval on 3.11.
- Delete setup.cfg: its [tool.black] section was written in INI syntax
  that black never reads; move the exclusion to pyproject.toml.
- splitinput.py: align the LineInfo.ofind() docstring with the warning
  (deprecated in 9.9, per the whatsnew entry for #15057).
- setupbase.py: drop a stale Python-2-era comment.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GEyCEHEuB6vN85jweKvQU7

- IPCompleter.limit_to__all__ config option (deprecated 5.0): completion
  on object.<tab> now always uses dir()-based discovery.
- IPCompleter.python_matches (deprecated 8.27): superseded by
  python_matcher, and was no longer part of the active matcher set.
- OInfo.get() (deprecated 8.13, transitional helper from when OInfo
  stopped being a dict in 8.12): access dataclass fields directly.
- pylabtools module-level backends/backend2gui __getattr__ shim
  (deprecated 8.24). The underlying dicts remain: they still back the
  find_gui_and_backend fallback for Matplotlib < 3.9.
- run_cell_async/should_run_async now raise TypeError instead of
  transforming the cell themselves when transformed_cell is not passed
  (DeprecationWarning since 7.17; ipykernel >= 6 always passes it).
  run_cell is unaffected.
- test_oinspect no longer needs the decorator package: the find_file
  wrapper-chain tests use functools.wraps, which sets the same
  __wrapped__ chain.

Completer.greedy (deprecated 8.8) was evaluated for removal and kept:
downstream projects (pyflyby's test suite, run in this repo's downstream
CI) still set it via %config, and removing it makes traitlets print
"Config option not recognized", breaking their expected-output tests.

The 8.6-era matcher-v1 methods (magic_config_matches,
python_func_kw_matches, dict_key_matches, dispatch_custom_completer)
were deliberately kept too: they are the implementations their v2
matcher wrappers delegate to.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GEyCEHEuB6vN85jweKvQU7

While investigating making IPython.embed/InteractiveShell/Application
lazy top-level attributes (reverted - see below), found a latent cycle
inside the terminal package: interactiveshell -> debugger -> embed ->
interactiveshell. It only worked because IPython/__init__.py happened
to import terminal.embed before anything reached terminal.debugger;
removing that ordering constraint made it a real ImportError.

terminal/debugger.py now imports terminal.embed inside do_interact,
the only place it's used, instead of at module level.

The lazy-attribute idea itself doesn't ship here: pyflyby's downstream
CI (green on main) broke because its _interactive.py does `import
IPython` and then accesses attribute chains like
IPython.terminal.ipapp.TerminalIPythonApp and
IPython.core.application.BaseIPythonApplication, which only resolve
through these imports' transitive side effects (its except
AttributeError path turns a missing attribute into a hard RuntimeError,
so `py` dies with EOF in their interactive tests). Left a comment
explaining the constraint; other downstreams likely have the same
pattern. A patch making pyflyby import those submodules explicitly has
been prepared for upstream, after which this is worth revisiting -
warm import time would drop from ~270ms to ~17ms.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GEyCEHEuB6vN85jweKvQU7

Summary of this cleanup batch: the deprecated APIs removed, the
modernization changes, and the performance fixes.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GEyCEHEuB6vN85jweKvQU7