Fix quantile boundaries and stale Rcpp exports

[xjsun1221]

Problem

This PR addresses two correctness and maintenance issues:

1. The sparse-probability path in Quantile() selects the wrong order
   statistic for interpolation. Both algorithm paths can also read one past
   the vector when probs contains 1.
2. The package uses hand-written native wrappers and init.c registration,
   while 22 functions still carry Rcpp::export attributes. Running
   compileAttributes() creates alternate wrappers, including a generated
   benchmark(exprs, env, tim, indices) function that conflicts with the
   public benchmark(..., times, envir, order) API.

Change

• Select the correct lower order statistic and guard exact/end-point
  quantiles in both algorithm paths.
• Remove a redundant partial selection before the dense path's full sort.
• Add regression coverage for unsorted probabilities, interpolation, and the
  0 and 1 boundaries.
• Remove stale Rcpp export attributes while preserving the existing wrappers,
  native registration, and public R API.

Validation

• Windows 11 x64
• R 4.5.0
• Rtools45 / GCC 14.2.0
• Rcpp::compileAttributes() completes without generating wrapper files.
• A clean source build, installation, and namespace load complete successfully.
• Quantile regression tests match stats::quantile(type = 7) in both the
  sparse- and dense-probability paths, including both endpoints.