Expand CUDA intrinsics coverage to 120+ operations

[mivertowski]

This commit significantly expands the CUDA codegen transpiler's intrinsic coverage from 45+ to 120+ GPU intrinsics across 13 categories:

New Intrinsics Added:

Synchronization (3 new)

• sync_threads_count, sync_threads_and, sync_threads_or

Atomic Operations (5 new)

• atomic_and, atomic_or, atomic_xor, atomic_inc, atomic_dec

Math Functions (8 new)

• trunc, fmod, remainder, copysign, cbrt, hypot, plus warp_size

Trigonometric (7 new)

• asin, acos, atan, atan2, sincos, sinpi, cospi

Hyperbolic (6 new)

• sinh, cosh, tanh, asinh, acosh, atanh

Exponential/Logarithmic (14 new)

• exp2, exp10, expm1, log2, log10, log1p, ldexp, scalbn, ilogb
• lgamma, tgamma, erf, erfc, erfinv, erfcinv

Classification (8 new)

• isnan, isinf, isfinite, isnormal, signbit, nextafter, fdim, nan

Warp Operations (8 new)

• warp_match_any, warp_match_all
• warp_reduce_add/min/max/and/or/xor

Bit Manipulation (8 new)

• popc, clz, ctz, ffs, brev, byte_perm
• funnel_shift_left, funnel_shift_right

Memory Operations (3 new)

• ldg, prefetch_l1, prefetch_l2

Special Functions (13 new)

• rcp, fdividef, saturate
• j0, j1, jn, y0, y1, yn (Bessel functions)
• normcdf, normcdfinv, cyl_bessel_i0, cyl_bessel_i1

Clock/Timing (3 new)

• clock, clock64, nanosleep

3D Stencil Support

• Added pos.up(buf) and pos.down(buf) for 3D volumetric kernels
• Full 3D at() offset support: pos.at(buf, dx, dy, dz)

DSL Module

• Comprehensive CPU fallback implementations for all intrinsics
• 20+ new tests for DSL functions

Testing

• 171 total tests (up from 143)
• New test coverage for all intrinsic categories
• Tests for intrinsic flags, categories, and CUDA output

Documentation

• Updated README with complete intrinsic reference
• Updated CLAUDE.md with new capabilities