Closes #5219: improve astype in ak.pandas.extension

[ajpotts]

Improve astype semantics across Arkouda pandas ExtensionArrays

Summary

This PR implements fully-featured, pandas-compatible astype behavior for all Arkouda-backed pandas ExtensionArrays:

• ArkoudaArray
• ArkoudaCategoricalArray
• ArkoudaStringArray

The implementation aligns with pandas’ ExtensionArray.astype contract, avoids unnecessary NumPy fallbacks, and consistently returns Arkouda-backed ExtensionArrays whenever possible.

In addition, this PR:

• Adds comprehensive type hints and overloads to satisfy mypy
• Expands string dtype normalization ("string" support)
• Introduces extensive unit tests covering numeric, string, categorical, object, and ExtensionDtype casting
• Makes doctests resilient to platform-dependent string-width differences

---

Key Changes

✅ Correct, consistent astype behavior

• object → always returns NumPy
• Same dtype + copy=False → returns self
• Numeric ↔ numeric → server-side Arkouda cast
• Categorical
  • category → stays categorical
  • string → ArkoudaStringArray
  • other dtypes → labels cast via Arkouda, returned as ExtensionArray
• Strings
  • string targets → stay ArkoudaStringArray
  • numeric/bool targets → server-side cast, return ExtensionArray
  • invalid numeric parses raise RuntimeError (documented and tested)

🧠 Pandas compatibility & typing

• Adds explicit @overload signatures matching pandas ExtensionArray.astype
• Fixes mypy override and return-type errors

🧪 Tests

New test coverage includes:

• Numeric → numeric casts
• ExtensionDtype targets (pd.Int64Dtype, pd.StringDtype, etc.)
• Categorical → string / numeric
• String → numeric / object
• Invalid string-to-numeric casts
• Copy vs no-copy semantics
• Doctest validation using ellipsis to avoid brittle unicode-width assertions

---

Why this matters

• Enables Arkouda ExtensionArrays to behave predictably inside pandas pipelines
• Avoids silent NumPy fallbacks that break distributed execution
• Clarifies and documents Arkouda’s server-side casting semantics
• Unblocks downstream pandas operations that rely on astype

Closes #5219: improve astype in ak.pandas.extension