improve builder pattern according to #20

[tchamelot]

Hey, this is a proposition to improve the builder pattern. It does not break the previous building method.

Now, one is able to build an instance with

let mut mpu9250 = MpuConfig::dmp().dmp_rate(DmpRate::_20Hz).build(i2c);
mpu9250.init(&mut Delay, &dmp_firmware).unwrap();

or

let mut mpu9250 = MpuConfig::imu().build(spi, ncs);
mpu9250.init(&mut Delay).unwrap();

If you have any suggestion, I will be glad to ear about it!

[little-arhat]

LGTM!

It would be nice to maybe update dmp example to show how to use new builders.

FYI, this branch increases size of a mpu code, compiled in release mode by 100bytes (according to cargo bloat):
with status quo version:
0.1% 1.3% 538B mpu9250

with this branch:
0.1% 1.6% 686B mpu9250

Not a huge increase, but maybe worth tracking that down to confirm increase is coming from useful functions.

Thanks!

[little-arhat]

For SPI connected devices we also provide *_with_reinit family of constructors. Although they are a bit ugly, they serve practical need -- maybe worth adding support for reinit to builders?

[tchamelot]

Could you give me an example of how you would use *with_reinit() so I can try to design something ergonomic.

[tchamelot]

FYI, this branch increases size of a mpu code, compiled in release mode by 100bytes (according to cargo bloat):
with status quo version:
0.1% 1.3% 538B mpu9250

with this branch:
0.1% 1.6% 686B mpu9250

Not a huge increase, but maybe worth tracking that down to confirm increase is coming from useful functions.

I did not know about cargo bloat, so i gave it a try on the rpi example.

master	builder legacy init	builder new init
0.2% 2.0% 4.8KiB mpu9250	0.2% 2.0% 4.8KiB mpu9250	0.2% 2.0% 4.9KiB mpu9250

First of all, the size of the mpu9250 crate is much bigger in my case. I do not see any change between this branch and the master when using the legacy init system. However, there is a 0.1KiB increase when using the new init system based on the builder pattern.

I can add it in the readme or in the doc if you think that such a small amount of memory should be documented.