Fall back from unsafe_trunc in UniformStep kernel for non-float ratio types

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas. Draft — opened by an agent.

Problem

A uniform Rational vector crashes on every search:

v = Rational{Int}[i//4 for i in 0:4:160]
searchsorted_last(Auto(v), v, 5//1)
# ERROR: MethodError: no method matching unsafe_trunc(::Type{Int64}, ::Rational{Int64})

Auto resolves a uniform vector to KIND_UNIFORM_STEP, and the closed-form kernels compute the index offset with unsafe_trunc(Int, floor(f)) / unsafe_trunc(Int, ceil(f)). unsafe_trunc is only defined for hardware floats and BigFloat; for a Rational ratio type it has no method. This is a pre-existing bug on main, independent of AD — found while testing SearchProperties/search robustness across element types (Float16/32/64, BigFloat, Rational, BigInt, Dual).

Fix

f is already clamped to the in-range interval before truncation, so the raw unsafe_trunc is replaced by a small dispatch helper:

@inline _uniform_floor_index(f::Base.IEEEFloat) = unsafe_trunc(Int, floor(f))
@inline _uniform_floor_index(f) = floor(Int, f)
@inline _uniform_ceil_index(f::Base.IEEEFloat) = unsafe_trunc(Int, ceil(f))
@inline _uniform_ceil_index(f) = ceil(Int, f)

• Base.IEEEFloat hot path keeps unsafe_trunc (in-range, so its range check is redundant) — unchanged.
• Every other ordered Real (Rational, BigFloat, AD Dual, …) falls back to the checked floor(Int, ·)/ceil(Int, ·), which those types support and which is exact for the clamped f.

Tests

Adds a UniformStep regression testset over uniform Rational{Int}, Rational{BigInt}, and BigFloat vectors, checking search results (exact-knot, between-knot, out-of-range) against Base.

• Full GROUP=Core suite: 172,183 pass locally.
• Runic clean.

Independence

Independent of the AD probe PR (#100) — touches src/kernels.jl and a disjoint region of the test file, no shared hunks, no ForwardDiff dependency. Either can merge first.