add isForkedChild() for tbb fork-safety checks

[kevinushey]

Summary

• Adds RcppParallel::isForkedChild() (C++) and isForkedChild() (R), returning TRUE when the current process is a fork() of the process in which RcppParallel (and TBB) was originally loaded.
• Captures getpid() once in R_init_RcppParallel; the C++ surface is a non-inline function in src/init.cpp, declared in inst/include/RcppParallel/Fork.h (matches the existing tbbParallelFor pattern, resolved at runtime via RTLD_GLOBAL on Linux/macOS, linker-imported on Windows).
• No-op on Windows (always returns FALSE) since Windows has no fork().

Closes #243.

Why this lives in RcppParallel

Downstream packages can't implement this themselves: RcppParallel may be loaded into the parent process long before the downstream package, so any baseline PID a downstream captures (e.g. in its own static initializer, or .onLoad) can already be a post-fork PID — exactly the false negative reproduced in the issue's library(RcppParallel); mclapply({ library(qs2); ... }) case. The baseline has to be captured by RcppParallel itself.

Suggested usage

#include <RcppParallel.h>

void doWork() {
  if (RcppParallel::isForkedChild()) {
    // serial fallback
    return;
  }
  RcppParallel::parallelFor(...);
}

Test plan

• R CMD INSTALL succeeds on macOS
• R: parent returns FALSE; both children in parallel::mclapply(1:2, ...) return TRUE
• C++: same result via Rcpp::sourceCpp calling RcppParallel::isForkedChild()
• CI: Linux + Windows builds

[kevinushey]

@traversc Let me know if this looks sufficient?

[traversc]

Looks good!

You might consider a different function name since it conflicts with parallelly::isForkedChild and has slightly different meaning.