add fenv cache to task struct (#51288)

Fixes #51277, though we give no guarantee that it keeps working this
way, or that calling `setrounding_raw` won't lead to other undefined
behavior.

To give some examples:

```julia
julia> t = Base.Rounding.setrounding_raw(Float64, Base.Rounding.to_fenv(RoundDown)) do
           Task(() -> println(rounding(Float64)))
       end
Task (runnable) @0x000000010dff04c0

julia> rounding(Float64)
RoundingMode{:Nearest}()

julia> wait(schedule(t))
RoundingMode{:Down}() # currently gives RoundingMode{:Nearest}()

julia> rounding(Float64)
RoundingMode{:Nearest}()

julia> Base.Rounding.setrounding_raw(Float64, Base.Rounding.to_fenv(RoundDown)) do
           Threads.@threads :static for i = 1:Threads.nthreads()
               println(Threads.threadid() =>  rounding(Float64))
           end
       end
1 => RoundingMode{:Down}()
2 => RoundingMode{:Down}() # currently gives RoundingMode{:Nearest}()
4 => RoundingMode{:Down}() # currently gives RoundingMode{:Nearest}()
3 => RoundingMode{:Down}() # currently gives RoundingMode{:Nearest}()
```

Author: simonbyrne

src/julia.h

@@ -20,6 +20,7 @@
 #include "libsupport.h"
 #include <stdint.h>
 #include <string.h>
+#include <fenv.h>
 
 #include "htable.h"
 #include "arraylist.h"
@@ -2282,6 +2283,9 @@ typedef struct _jl_task_t {
     uint16_t priority;
 
 // hidden state:
+    // cached floating point environment
+    // only updated at task switch
+    fenv_t fenv;
 
     // id of owning thread - does not need to be defined until the task runs
     _Atomic(int16_t) tid;

src/task.c

@@ -534,6 +534,7 @@ JL_NO_ASAN static void ctx_switch(jl_task_t *lastt)
     jl_set_pgcstack(&t->gcstack);
     jl_signal_fence();
     lastt->ptls = NULL;
+    fegetenv(&lastt->fenv);
 #ifdef MIGRATE_TASKS
     ptls->previous_task = lastt;
 #endif
@@ -726,6 +727,7 @@ JL_DLLEXPORT void jl_switch(void) JL_NOTSAFEPOINT_LEAVE JL_NOTSAFEPOINT_ENTER
            0 == ptls->finalizers_inhibited);
     ptls->finalizers_inhibited = finalizers_inhibited;
     jl_timing_block_task_enter(ct, ptls, blk); (void)blk;
+    fesetenv(&ct->fenv);
 
     sig_atomic_t other_defer_signal = ptls->defer_signal;
     ptls->defer_signal = defer_signal;
@@ -1138,6 +1140,7 @@ JL_DLLEXPORT jl_task_t *jl_new_task(jl_function_t *start, jl_value_t *completion
     t->excstack = NULL;
     t->ctx.started = 0;
     t->priority = 0;
+    fegetenv(&t->fenv);
     jl_atomic_store_relaxed(&t->tid, -1);
     t->threadpoolid = ct->threadpoolid;
     t->ptls = NULL;
@@ -1239,6 +1242,7 @@ CFI_NORETURN
     if (!pt->sticky && !pt->ctx.copy_stack)
         jl_atomic_store_release(&pt->tid, -1);
 #endif
+    fesetenv(&ct->fenv);
 
     ct->ctx.started = 1;
     JL_PROBE_RT_START_TASK(ct);