Add 2.0 deprecation warnings for 1.4 second-wave removals (#263)

[derek73]

Summary

One themed PR, mirroring how the 1.3.0 sweep shipped (#253), so every decided 2.0 removal in the second wave warns ahead of time. Closes #263; the parent 2.0-removal issues (#255, #256, #258, #245, #279) stay open for their actual removal halves.

• empty_attribute_default assignment (Remove empty_attribute_default; empty attributes always return '' #255): once None support goes, the only legal value left is the default -- a dial with one position isn't configuration. Assignment now emits DeprecationWarning via a new _EmptyAttributeDefaultAttribute descriptor and rejects non-str/non-None values outright with TypeError; reading the attribute is unaffected.
• TupleManager/RegexTupleManager unknown-key attribute access (TupleManager unknown-key access silently returns None / EMPTY_REGEX; raise AttributeError in 2.0 #256): a misspelled or omitted key (CONSTANTS.regexes.parenthesys, CONSTANTS.capitalization_exceptions.phd_typo) currently degrades silently (None/EMPTY_REGEX) with no traceback pointing at the typo. Now warns naming the miss and the known keys; .get() remains available for intentional soft access. Single-underscore introspection probes (IPython's _repr_html_, _ipython_canary_method_should_not_exist_, etc.) are excluded alongside the existing dunder guard, so displaying CONSTANTS.regexes in a notebook doesn't misfire the warning.
• HumanName slice __getitem__ and all of __setitem__ (Deprecate slice __getitem__ and __setitem__ on HumanName #258): field access by position (name[1:-3]) has no real use case, and item assignment (name['first'] = x) duplicates plain attribute assignment. Both now warn; string-key access (name['first']) is unaffected. docs/usage.rst's quick-start doctest drops the now-deprecated slicing/item-assignment examples -- string-key access was already demonstrated there via name["title"].
• SetManager.add_with_encoding() (Drop bytes input in 2.0; require str #245): the method itself now warns on every call regardless of argument type, naming add() as the replacement. Previously only the bytes-decode path warned, so the method's own removal was silently unwarned on the str path.
• Constants.__setstate__'s legacy-pickle migration shim (Constants.__setstate__ loads pre-1.3.0 pickle blobs via a silent migration shim; drop the shim in 2.0 #279): warns once per __setstate__ call (not once per skipped key, verified with two stale keys in one blob) when it loads a pre-1.3.0 blob carrying a stale computed-property key, telling users to re-pickle.

Notable implementation detail

__getstate__/__setstate__ now use inspect.getattr_static instead of getattr to detect descriptor types without triggering __get__ -- needed because the new _EmptyAttributeDefaultAttribute.__get__(obj=None, ...) intentionally returns '' (not self) to preserve the pre-existing str-only mypy inference on empty_attribute_default (matches the deliberate typing choice from PR #250). That same ''-not-self choice is also load-bearing for Constants.__repr__'s default-value comparison, not just mypy -- documented inline. Restoring pickled/copied state bypasses the descriptor's warning-emitting setter, since that isn't a user assignment.

Test plan

• TDD throughout: every new warning path had a failing test before implementation
• Suite verified noise-free, including under -W error::DeprecationWarning (1540 passed / 22 xfailed, zero warnings beyond one pre-existing unrelated marker warning)
• mypy clean, ruff clean, Sphinx doctest + html builds clean
• Two independent multi-agent review passes (code, tests, comments, silent-failures, type design) caught and fixed real gaps before landing: a sunder-probe false positive on TupleManager unknown-key access silently returns None / EMPTY_REGEX; raise AttributeError in 2.0 #256; an overly-broad warning-suppression scope in tests/conftest.py's autouse fixture that would have silently swallowed any future deprecation warning on the other 7 scalar config attrs; missing type validation on empty_attribute_default; a stale "~8 properties" comment; and three test-coverage gaps (Constants.copy() preserving a customized empty_attribute_default, an exact-warning-count check on add_with_encoding()'s str path, and a genuine once-per-call-not-per-key check for Constants.__setstate__ loads pre-1.3.0 pickle blobs via a silent migration shim; drop the shim in 2.0 #279)
• docs/release_log.rst updated under 1.4.0 Unreleased; docs/usage.rst's quick-start doctest trimmed of the deprecated/redundant slicing, len(), list(), and item-access examples

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.67%. Comparing base (4aaedb8) to head (afb5571).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #286      +/-   ##
==========================================
- Coverage   97.70%   97.67%   -0.04%     
==========================================
  Files          13       13              
  Lines         917      947      +30     
==========================================
+ Hits          896      925      +29     
- Misses         21       22       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.