[Code scan] Replace np.ndarray constructors used for coordinate data conversion

This issue comes from a Codex global scan of deepmodeling/tbplas at commit 4d3652b3dba8eb930e4869a6d41074f62d703692.

Severity: Medium

Two input-conversion branches use `np.ndarray(...)` where they should use `np.array(...)`. `np.ndarray` treats the argument as a shape, not data: list inputs raise `TypeError`, while tuple inputs can allocate uninitialized coordinates. That breaks normal list/tuple inputs and can silently produce invalid results.

Code references:
https://github.com/deepmodeling/tbplas/blob/4d3652b3dba8eb930e4869a6d41074f62d703692/tbplas/base/lattice.py#L116-L120
https://github.com/deepmodeling/tbplas/blob/4d3652b3dba8eb930e4869a6d41074f62d703692/tbplas/diagonal/lindhard.py#L541-L542

Suggested fix: replace both constructors with `np.array(..., dtype=...)`, matching the surrounding conversion helpers.

Start at tbplas/base/lattice.py lines 116-120 and tbplas/diagonal/lindhard.py lines 541-542, then compare the surrounding conversion helpers. Verify list and tuple coordinate inputs in both branches convert as data with the intended dtype, without raising errors or producing uninitialized values.

Written by the indexing model from the issue text.