Wire calculatePartialSums to native SIMD via Panama FFI downcall

[r-devulap]

This change uses a native implementation of calculatePartialSums to accelerate PQ query scoring.
On ada002-100k with FUSED_PQ (numPQsubspaces/M =96, JDK build 23.0.1+11-39), it delivers 2–3× higher QPS and 40–65% lower mean latency across common overquery settings. Index build time, disk usage, and heap usage show no meaningful regression. The optimization is isolated to the PQ path; non‑PQ queries are unaffected.

Combined QPS and Latency Results (FUSED_PQ)

topK = 10

Overquery	QPS (main)	QPS (native)	Speedup	Latency ms (main)	Latency ms (native)	Latency ↓
1×	8,987	26,101	2.9×	0.462	0.171	−63%
2×	8,361	21,706	2.6×	0.505	0.202	−60%
5×	7,199	14,364	2.0×	0.590	0.292	−51%
10×	5,796	9,640	1.7×	0.731	0.431	−41%

topK = 100

Overquery	QPS (main)	QPS (native)	Speedup	Latency ms (main)	Latency ms (native)	Latency ↓
1×	5,743	9,568	1.7×	0.736	0.439	−40%
2×	4,174	5,574	1.3×	1.022	0.733	−28%
``

Summary of changes in this PR:

• Wire calculatePartialSums in NativeVectorUtilSupport to a new Panama FFI downcall for the native calculate_partial_sums_f32_512 SIMD implementation.
• Replace icelake-server gcc target with skylake-avx512 in build script (icelake-server isnt required to build our native code)
• Remove global mutable state: eliminate initialIndexRegister, indexIncrement, maskSeventhBit, maskEighthBit globals and their constructor initializer; move mask constants (maskSeventhBit, maskEighthBit) to local scope inside lookup_partial_sums
• Add shared reduce_add_128_ps and reduce_add_256_ps helper functions using proper horizontal-add sequences instead of store-to-array loops
• Remove redundant if (length >= N) guards in all SIMD kernels — the loop body already handles the zero-iteration case correctly
• Replace store-to-aligned-array horizontal reduction pattern with the new helpers across all 128- and 256-bit dot product and euclidean distance functions
• Remove preferred_size parameter from dot_product_f32 and euclidean_f32; always dispatch to AVX-512 when length >= 16
• Standardize inline annotations: replace attribute((always_inline)) inline with JV_FINLINE / JV_INLINE macros throughout

[github-actions]

Before you submit for review:

• Does your PR follow guidelines from CONTRIBUTIONS.md?
• Did you summarize what this PR does clearly and concisely?
• Did you include performance data for changes which may be performance impacting?
• Did you include useful docs for any user-facing changes or features?
• Did you include useful javadocs for developer oriented changes, explaining new concepts or key changes?
• Did you trigger and review regression testing results against the base branch via Run Bench Main?
• Did you adhere to the code formatting guidelines (TBD)
• Did you group your changes for easy review, providing meaningful descriptions for each commit?
• Did you ensure that all files contain the correct copyright header?

If you did not complete any of these, then please explain below.