Promote mixed-type list literals like NumPy (#539)

[ciaranra]

What

Fixes #539. pecos.array([0, 1.5, True]) produced a bool array, silently destroying the 1.5. Root cause: the generic inference path had a correct promotion lattice for int -> float -> complex, but bool was wired backwards -- promote_type_to_bool DEMOTED accumulated integers and floats into != 0 bools whenever a Python bool appeared. In NumPy's lattice, bool is the bottom: it never drags a numeric type down.

The fix inverts the flow and deletes the demoter:

• A bool claims the dtype only while nothing has been accumulated (so [True, False] is still bool); otherwise it is absorbed into the already-promoted type as 0/1.
• Integers, floats, and u64-range values arriving after bools lift the accumulated bools upward (promote_bools_to_i64 / promote_bools_to_f64; the complex direction already existed and was correct).
• Mixing bools with Pauli/PauliString literals now raises a clear TypeError (previously an internal shape error).

The #540 homogeneous fast paths are untouched -- they bail to this generic path on any mixed list, which is exactly where the defect lived.

Verification

• New oracle tests: 11 mixed-literal cases asserting dtype and values against the installed NumPy, both element orders, including [True, 2**63] -> uint64, complex both directions, and nested [[1, True], [2.5, False]]; plus the Pauli-mixing TypeError.
• Full pecos-rslib suite: 2,124 pass. Downstream quantum-pecos suite: 4,764 pass.
• Cold cargo clippy -p pecos-rslib --all-targets -- -D warnings, cargo fmt --check, unshielded pre-commit run --all-files: clean.