Proposal: _fast variants for lower-overhead Julia calls

The current overhead for me calling julia_call("sqrt", 2) is ~7μs. Providing a _fast equivalent stripping it of some QoL stuff reduces this overhead to ~4.5μs for me. Here is the code I added:

julia_do.call_fast <- julia$do.call_fast <- function(func_name, arg_list) {
    r <- .julia$do.call_(list(fname = func_name, args = arg_list, need_return = "R", show_value = FALSE))
    if (inherits(r, "error")) stop(r)
    r
}

julia_call_fast <- julia$call_fast <- function(func_name, ...)
    julia$do.call_fast(func_name, list(...))

And benchmark:

bench::mark(julia_call("sqrt", 2), julia_call_fast("sqrt", 2))[, 1:3]
# A tibble: 2 × 3
  expression                          min   median
  <bch:expr>                     <bch:tm> <bch:tm>
1 "julia_call(\"sqrt\", 2)"         6.4µs    7.0µs
2 "julia_call_fast(\"sqrt\", 2)"    4.1µs    4.5µs

Happy to make a PR if you want these added _fast equivalents. I also understand if saving ~2.5μs per call isn't worth the new API.

Start by locating the existing julia_call and julia$do.call_ entry points and compare their current overhead-related behavior with the proposed wrappers. Confirm the fast variants preserve the shown error handling and return behavior, then benchmark julia_call against julia_call_fast to verify the intended reduction.