Add np.may_share_memory / np.shares_memory / np.byte_bounds to _PASSTHROUGH_NAMES

[spMohanty]

Identified in #67 — flagged as cleanup in the "Explicitly out of scope" section.

Current state

_PASSTHROUGH_NAMES (in src/whest/_ndarray.py) holds the original numpy callables that should bypass __array_function__ entirely because they're zero-FLOP queries. Today the set covers shape / type queries:

_PASSTHROUGH_NAMES = (
    "ndim", "shape", "size",
    "result_type", "can_cast", "min_scalar_type",
    "promote_types", "find_common_type", "mintypecode",
    "array_equal",
)

Missing: may_share_memory, shares_memory, byte_bounds. They're zero-FLOP memory-layout queries that currently work because the protocol path strips inputs and dispatches; routing them through the strip path is wasteful.

Reproducer

import numpy as np, whest as we
A = we.random.randn(3, 3)
np.may_share_memory(A, A)   # works, but enters __array_function__ unnecessarily

Suggested fix

Add the three names to _PASSTHROUGH_NAMES and update the eager _INITIAL_PASSTHROUGH = _build_passthrough() capture. One-line addition each. Add a short test in tests/test_array_protocols.py that asserts these calls no longer enter __array_function__ (e.g. by patching _get_array_function_dispatch and asserting it's not called).

Acceptance criteria

• The three names added to _PASSTHROUGH_NAMES.
• Test pinning that they don't dispatch through __array_function__.

Related

• PR Track method calls and NumPy protocols on WhestArray (closes #58, #38, #62) #67 — introduced _PASSTHROUGH_NAMES and called this out as a cleanup follow-up.