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

Umbrella for the 1.4 half of the second-wave 2.0 removals, so the v1.4 milestone (and the release-log audit against it) sees this work — the parent issues live on the v2.0 milestone and only their removal halves belong there. One themed PR, mirroring how the 1.3.0 sweep shipped (#253).

Checklist (one warning per parent issue's "Bridge (1.4)" section)

• Remove empty_attribute_default; empty attributes always return '' #255 — DeprecationWarning on any assignment to empty_attribute_default (the __str__ mangling fix is tracked separately as str() strips literal "None" from real names when empty_attribute_default is None #254, already on v1.4)
• TupleManager unknown-key access silently returns None / EMPTY_REGEX; raise AttributeError in 2.0 #256 — DeprecationWarning on unknown-key attribute access on TupleManager/RegexTupleManager, naming the miss and the known keys
• Deprecate slice __getitem__ and __setitem__ on HumanName #258 — DeprecationWarning on slice __getitem__ and on __setitem__; replace the usage.rst slicing example
• Drop bytes input in 2.0; require str #245 — DeprecationWarning on any add_with_encoding() call (str included), naming add(); the shipped 1.3.0 warning only covers bytes reaching it
• Constants.__setstate__ loads pre-1.3.0 pickle blobs via a silent migration shim; drop the shim in 2.0 #279 — DeprecationWarning on legacy-format unpickle (pre-1.3.0 Constants blobs hitting the __setstate__ migration shim), once per __setstate__ call, telling users to re-pickle

Not in this sweep: the constants=None warning ships with Constants.copy() in #260 (they're one coherent change), and the first-wave warnings (#223/#243 items, and #245's bytes warnings) already shipped in 1.3.0.

Same standards as #253: TDD, warnings name their replacement and link the parent issue, suite noise-free including under -W error::DeprecationWarning.