Fix OpenMP data races and add speed improvements in Rcpp src/ functions

[Copilot]

Several Rcpp functions contained OpenMP data races producing undefined behavior under multi-threaded execution, plus avoidable redundant work in single-threaded hot paths.

Race condition fixes

• sum_segim.cpp (profoundSegimFlux): Threads wrote to shared fluxes[k] without synchronization. Replaced with thread-local reduction arrays merged via #pragma omp critical.
• this_in_that.cpp (vec_this_in_vec_that, mat_this_in_vec_that): Parallel writes of true to the same ref_ID index is UB. Removed #pragma omp from the lookup-table setup loop — it's O(N) and fast enough single-threaded.
• ellip_cover.cpp / aper_cover.cpp (profoundEllipWeight, profoundAperWeight): Overlapping sources processed on different threads would race on weight(i,j) +=. Added #pragma omp atomic on each accumulation (avoids the memory cost of per-thread weight matrices).

Performance improvements

• akima.cpp (interpolateLinearGrid): The y-bracket index search (top_index/bottom_index) was O(ncol_tiles) per output pixel, running redundantly inside the outer row loop. Hoisted to a precomputation pass outside the row loop — reduces total index-search work from O(rows × cols × tiles) → O(cols × tiles).

• skygrid.cpp (Cadacs_FindSkyCellValues): Replaced Rcpp::NumericVector(skyN) (zero-initializes all elements) + indexed fill with std::vector::reserve(skyN) + push_back, eliminating the zero-initialization pass.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[asgr]

Looks good, and I've learnt a bit about OpenMP races.