Optimize C ABI boundaries

[mappu]

The Go binding calls a C ABI binding, which calls the Qt C++ functions. Overall, it's somewhat efficient.

But to comply with the common C ABI, there are some places where data types must be converted, or moved onto the heap. This could be optimized in some cases.

Ideas:

• Bypass malloc/free for zero-length strings and arrays
• When returning a QString, have C++ fill in a Go-provided buffer, instead of copying the data twice
  • Current: C malloc - C strcpy - Cgo boundary - Go malloc - Go memcpy - Cgo boundary - C free - Cgo boundary - Go epilogue
  • Proposal: C calls go_malloc - Cgo boundary - Go malloc - Cgo boundary - C strcpy - Cgo boundary - Go epilogue
  • Same 3x boundaries

[mappu]

• Pass struct miqt_array by value instead of pointer (saves 1x malloc)
• Pass struct miqt_string by value instead of pointer (still one malloc, but simpler)