Speedup RGB565 rendering

[ogoffart]

By not going through a RGB888 premultiplied color, but using a special encoded format for premultiplied color so we can operate on all color at once in a u32

Unfortunately, this is a breaking change as it changes the TargetPixel trait.

[ogoffart]

Result of cargo bench -p printerdemo_mcu (with the memory mesure remoced)

Before

Timer precision: 73 ns
bench                fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ full                            │               │               │               │         │
│  ├─ Rgb565Pixel    2.281 ms      │ 4.479 ms      │ 3.244 ms      │ 3.251 ms      │ 100     │ 100
│  ╰─ Rgb<u8>        4.939 ms      │ 7.232 ms      │ 6.029 ms      │ 6.032 ms      │ 100     │ 100
╰─ render_only                     │               │               │               │         │
   ├─ Rgb565Pixel                  │               │               │               │         │
   │  ├─ FullBuffer  1.096 ms      │ 1.287 ms      │ 1.168 ms      │ 1.177 ms      │ 100     │ 100
   │  ╰─ LineByLine  1.085 ms      │ 1.297 ms      │ 1.191 ms      │ 1.189 ms      │ 100     │ 100
   ╰─ Rgb<u8>                      │               │               │               │         │
      ├─ FullBuffer  1.432 ms      │ 1.826 ms      │ 1.554 ms      │ 1.558 ms      │ 100     │ 100
      ╰─ LineByLine  1.478 ms      │ 1.818 ms      │ 1.573 ms      │ 1.594 ms      │ 100     │ 100

With this change

Timer precision: 49 ns
bench                fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ full                            │               │               │               │         │
│  ├─ Rgb565Pixel    1.989 ms      │ 4.086 ms      │ 2.932 ms      │ 2.972 ms      │ 100     │ 100
│  ╰─ Rgb<u8>        4.62 ms       │ 8.459 ms      │ 5.441 ms      │ 5.555 ms      │ 100     │ 100
╰─ render_only                     │               │               │               │         │
   ├─ Rgb565Pixel                  │               │               │               │         │
   │  ├─ FullBuffer  859.6 µs      │ 1.023 ms      │ 917.9 µs      │ 925.1 µs      │ 100     │ 100
   │  ╰─ LineByLine  828.9 µs      │ 1.048 ms      │ 941.3 µs      │ 956.4 µs      │ 100     │ 100
   ╰─ Rgb<u8>                      │               │               │               │         │
      ├─ FullBuffer  1.336 ms      │ 1.593 ms      │ 1.441 ms      │ 1.445 ms      │ 100     │ 100
      ╰─ LineByLine  1.392 ms      │ 1.618 ms      │ 1.517 ms      │ 1.496 ms      │ 100     │ 100

---

I don't understand why this also improve the RGB8 benchmark