Fix open-addressing index generation and wraparound

[sleeepyjack]

Closes #834
Closes #837

• make hash-to-index reduction defined across the supported hash and extent ranges
• avoid signed and unsigned overflow when advancing probing iterators
• reject capacities that cannot be represented after stride or prime rounding
• add regression coverage for scalar and cooperative-group linear/double probing and signed/unsigned extent boundaries

[sleeepyjack]

Benchmark results

Environment:

• GPU: NVIDIA RTX PRO 6000 Blackwell Max-Q, sm_120
• CUDA: 13.1.115
• Host compiler: GCC 14.3
• Build type: Release
• Input: 20 million int32_t keys
• Baseline: upstream/dev at 4b26118c
• Candidate: e6894e31

The main performance change is the probe iterator advancement. Previously, each collision advanced
the iterator using:

(current + step) % upper_bound

For dynamic extents, this invokes the fast_int division/modulo path on every additional probe.
The new implementation relies on the iterator invariants (current < upper_bound and
step <= upper_bound) and performs at most one wrap using a comparison and add/subtract.

As a result, the improvement scales with probe-chain length:

• Higher occupancies produce more collisions and therefore larger improvements.
• Non-matching lookups traverse longer probe sequences than successful lookups and benefit most.
• Linear probing shows the largest gains at high occupancy because clustering creates long chains.
• Insert gains are smaller because successful inserts often terminate earlier and include atomic
  insertion costs that are unaffected by this change.
• Capacity validation runs on the host during construction and does not affect timed kernels.

Across the focused matrix:

Configuration group	Cases	Faster	Unchanged	Slower
Default extent, double hashing	15	14	1	0
extent<int32_t>, double hashing	21	21	0	0
extent<int32_t>, linear probing	12	9	3	0
Total	48	44	4	0

Default extent: contains by matching rate

Occupancy is fixed at 0.8.

Key	Distribution	NumInputs	Occupancy	MatchingRate	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
I32	UNIQUE	20000000	0.8	0	1.377 ms	0.42%	1.235 ms	0.32%	-141.887 us	-10.30%	FAST
I32	UNIQUE	20000000	0.8	0.5	1.140 ms	0.38%	1.027 ms	0.29%	-113.240 us	-9.93%	FAST
I32	UNIQUE	20000000	0.8	1	826.834 us	0.48%	753.128 us	0.44%	-73.706 us	-8.91%	FAST

Default extent: find by matching rate

Occupancy is fixed at 0.8.

Key	Distribution	NumInputs	Occupancy	MatchingRate	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
I32	UNIQUE	20000000	0.8	0	1.442 ms	0.50%	1.275 ms	0.40%	-166.867 us	-11.57%	FAST
I32	UNIQUE	20000000	0.8	0.5	1.261 ms	0.35%	1.137 ms	0.45%	-124.080 us	-9.84%	FAST
I32	UNIQUE	20000000	0.8	1	963.968 us	0.36%	890.785 us	0.34%	-73.183 us	-7.59%	FAST

Default extent: insert by occupancy

Key	Distribution	NumInputs	Occupancy	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
I32	UNIQUE	20000000	0.9	1.577 ms	0.33%	1.486 ms	0.42%	-90.445 us	-5.74%	FAST
I32	UNIQUE	20000000	0.8	1.536 ms	1.48%	1.481 ms	1.12%	-54.285 us	-3.53%	FAST
I32	UNIQUE	20000000	0.5	2.829 ms	0.06%	2.828 ms	0.06%	-1.494 us	-0.05%	SAME

Signed extent, double hashing: contains

NumInputs	Occupancy	MatchingRate	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
20000000	0.9	0.5	1.502 ms	0.41%	1.406 ms	0.25%	-96.631 us	-6.43%	FAST
20000000	0.8	0.5	1.063 ms	0.54%	994.997 us	0.29%	-67.614 us	-6.36%	FAST
20000000	0.9	0	1.966 ms	0.40%	1.849 ms	0.19%	-117.584 us	-5.98%	FAST
20000000	0.9	1	858.481 us	0.44%	809.869 us	0.37%	-48.612 us	-5.66%	FAST
20000000	0.8	0	1.269 ms	0.44%	1.199 ms	0.29%	-70.282 us	-5.54%	FAST
20000000	0.8	1	774.082 us	0.50%	732.542 us	0.50%	-41.540 us	-5.37%	FAST
20000000	0.5	0	1.017 ms	2.13%	987.446 us	1.46%	-29.280 us	-2.88%	FAST
20000000	0.5	0.5	954.535 us	0.25%	928.963 us	0.39%	-25.571 us	-2.68%	FAST
20000000	0.5	1	863.941 us	0.19%	851.916 us	0.22%	-12.025 us	-1.39%	FAST

Signed extent, linear probing: representative high-occupancy cases

Operation	Occupancy	MatchingRate	Ref Time	Cmp Time	%Diff
Contains	0.9	0.5	3.985 ms	3.495 ms	-12.28%
Contains	0.9	0	5.084 ms	4.469 ms	-12.10%
Contains	0.8	0	1.366 ms	1.211 ms	-11.35%
Contains	0.8	0.5	1.119 ms	993.131 us	-11.23%
Insert	0.9	-	1.526 ms	1.450 ms	-4.97%