fix: normalize negative integer slice bounds in indexing

[henryiii]

🤖 AI text below 🤖

Brings NumPy-style negative and out-of-range slice indexing to boost-histogram, fixing the bug at the underlying layer (mirrors scikit-hep/hist#699, so hist and other UHI consumers get it for free).

Negative bounds

Negative slice bounds raised ValueError: begin < end required (e.g. h[1:-2]) because plain negative ints were passed straight to Boost.Histogram, which does not apply NumPy-style negative indexing. Axis._process_callable now normalizes plain negative ints relative to the axis length. Locators (bh.loc, bh.tag.at, rebin/sum steps) are callables handled earlier and stay untouched, so flow access like tag.at(-1) still reaches underflow. Scalar negative indexing (h[-1]) already worked via _compute_uhi_index.

Out-of-range bounds (clamping)

Out-of-range bounds now clamp to [0, len] like NumPy, so every non-empty slice matches NumPy exactly:

slice	result
h[5:100]	h[5:]
h[:100]	h[:]
h[-15:5]	h[:5]
h[-15:]	h[:]

Empty slices

NumPy returns an empty array for slices like h[1:-15] or h[5:2], but a Boost.Histogram Regular/Variable axis cannot have zero bins. These now raise a clear IndexError explaining the limitation instead of the cryptic begin < end required / bins > 0 required. The check lives in _handle_slice, so intentional flow access via integration — e.g. h[3::sum] summing the overflow on a size-3 axis — is preserved.

Tests

tests/test_histogram_indexing.py:

• test_negative_index_access — negative scalars, slices, one-sided, dict-form, per-axis length on 2D.
• test_out_of_range_slice_clamps_like_numpy — extents matched against np.arange for clamped cases.
• test_empty_slice_raises — IndexError for the empty cases, including reversed h[5:2:sum].

One pre-existing assertion in test_single_flow_bin was updated: h[3::sum] on a 3-category axis (start past the last bin) now raises the clearer IndexError instead of ValueError.

Verified slice extents match numpy across negative/out-of-range/empty cases. Full suite: 1127 passed, 1 xfailed.

Note on the Hist 🐍 3.14 CI job: any failure there is unrelated — the downstream hist test suite passes; only its trailing mypy step trips on a numpy-stubs/Python-3.14 version mismatch (numpy/__init__.pyi: Type statement is only supported in Python 3.12 and greater).