Fix very slow emulation speed under Emscripten

The use of setjmp/longjmp to handle PPC exceptions results in WASM <-> JS
jumps in generated Emscripten code. This should be fine (because there are
not a lot of exceptions), but for reasons that I don't fully understand,
we were crossing that boundary for every single instruction.

Work around this by not allowing ppc_exec_inner to be inlined, which adds
another function on the callstack and appears to keep the execution
entirely on the WASM side.

Timing of booting the 7.1.2 Disk Tools floppy:

                             Before     After
Happy Mac                    43s        17s
Welcome to Power Macintosh   62s        24s
Desktop Appears              100s       38s
Finder is running            192s       70s

Author: mihaip

cpu/ppc/ppcexec.cpp

@@ -321,7 +321,14 @@ void force_cycle_counter_reload()
 
 /** Execute PPC code as long as power is on. */
 // inner interpreter loop
-static void ppc_exec_inner()
+static
+// Don't inline this function under Emscripten, otherwise we will end up with
+// very inefficient code generation due to the setjmp call in the parent
+// ppc_exec() function.
+#ifdef EMSCRIPTEN
+__attribute__((noinline))
+#endif
+void ppc_exec_inner()
 {
     uint64_t max_cycles;
     uint32_t page_start, eb_start, eb_end;