cost_distance: replace mutable default target_values=[] with None sentinel (#3340)

[brendancol]

Closes #3340

What

cost_distance() declared target_values: list = [] (a mutable default). Switched it to the None sentinel with internal substitution, matching proximity() / allocation().

• Signature: target_values: list = None
• Body: if target_values is None: target_values = [] before the existing np.asarray rebind
• Regression tests pinning the None default and confirming None / [] / omitted are equivalent

Why

A shared [] default is the unsafe form and drifted from the sibling proximity functions. The value was already rebound via np.asarray on entry, so there was no active mutation bug, but the declaration is still wrong and inconsistent. Found by /sweep-api-consistency (Cat 4, MEDIUM).

Non-breaking: callers passing a list are unaffected and the default behavior (all non-zero finite pixels are sources) is identical. No deprecation shim needed since nothing is renamed.

Backends

Pure signature/default change in the public wrapper, shared by all four backends. Verified numpy, cupy, dask+numpy, dask+cupy still dispatch and produce identical results.

Test plan

• pytest xrspatial/tests/test_cost_distance.py (57 passed)
• New: default is None, and None / [] / omitted give equal output
• Existing test_target_values (explicit values still filter) passes on all parametrized backends

[brendancol]

PR Review: cost_distance — replace mutable default target_values=[] with None sentinel

Blockers

None.

Suggestions

None.

Nits

• xrspatial/cost_distance.py:1230 — the docstring still reads "If empty, all non-zero finite pixels are sources." With the default now None, "if not provided" would be marginally more precise (proximity's docstring phrases it that way). Optional; the optional marker on the type line already covers the None default.

What looks good

• Minimal, surgical change: signature default plus a two-line None->[] substitution placed before the existing np.asarray rebind. No behavior change.
• Matches the sibling convention exactly — proximity.py uses the same target_values: list = None plus internal substitution.
• Non-breaking, correctly identified as needing no deprecation shim (nothing renamed).
• Tests pin both the regression (default is None) and the behavioral equivalence of None / [] / omitted, with source-pixel assertions that confirm semantics rather than just "runs without error."
• Backend dispatch untouched; the change lives entirely in the shared public wrapper, so all four backends inherit it.

Checklist

• No algorithm change (signature/default only)
• All backends inherit the change (shared wrapper)
• NaN handling unchanged
• Edge cases covered (None / [] / omitted equivalence)
• No premature materialization or copies introduced
• Benchmark not needed
• README matrix not applicable
• Docstring present; minor wording nit noted

[brendancol]

Follow-up: applied the docstring nit (cost_distance.py:1230, "If empty" -> "If not provided") to match the proximity wording now that the default is the None sentinel. Tests still green. No blockers or suggestions outstanding.