Fix jl_new_task to avoid allocating stack twice

[fingolfin]

If the user requested dedicated stack of a certain size, then jl_new_task immediately allocates the stack by invoking jl_alloc_fiber.

However, it also later nulls t->stkbuf. As a result, ctx_switch invokes jl_alloc_fiber again to allocate another stack.

An alternate fix (assuming my explanation above is even correct...) might be to not pre-allocate the stack here, and leave that to ctx_switch.

Resolves #36361

[fingolfin]

Can this be merged? The failing tests seem to be spurious; at least I have seen them on plenty other PRs:

[ Info: 
backtrace                          (9) |     5.88 |   0.06 |  1.0 |      79.85 |   915.53
exceptions                         (9) |        started at 2021-01-07T16:34:08.497
spawn                              (5) |         failed at 2021-01-07T16:34:10.185
Error During Test at C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\test\testdefs.jl:21
  Got exception outside of a @test
  LoadError: Failed to connect to cache.julialang.org port 443: Timed out while requesting https://cache.julialang.org/https://frippery.org/files/busybox/busybox.exe

and

REPL                               (4) |         failed at 2021-01-08T00:36:15.070
Error During Test at /buildworker/worker/tester_linuxaarch64/build/share/julia/test/testdefs.jl:21
  Got exception outside of a @test
  LoadError: LoadError: "hard kill repl test"

[fingolfin]

buildbot/tester_linuxaarch64 still fails with the same error:

Error During Test at /buildworker/worker/tester_linuxaarch64/build/share/julia/test/testdefs.jl:21
  Got exception outside of a @test
  LoadError: LoadError: "hard kill repl test"
  Stacktrace:
    [1] try_yieldto(undo::typeof(Base.ensure_rescheduled))
      @ Base ./task.jl:705

The same appears in many, many other PRs. E.g. right now I see it in PRs #39358, #39312, #39287, #39285 (and I didn't go beyond the first page of open PRs)

[fingolfin]

Yay, buildbot/tester_linuxaarch64 seems to have passed now