BUG: fix incorrect code generation when a struct ends in a bitfield and --explicit-padding is used

[sw17ch]

A struct ending in a bitfield will have its trailing padding duplicated when --explicit-padding is used. This PR adds a test showing this condition, and performs a small check for whether or not the previous field was a bitfield before emitting trailing padding.

Please Consider

I do not know if my fix is a good fix, it is merely a fix. I wrote this at 1AM, and it may contain errors in judgement consistent with the time of night.

Also, I was able to find this bug because we run the bindgen_test_layout_* test cases in our CI system, and they successfully flagged the difference in size, so thank you for that!

Reproduction

Before this patch, the following header would have incorrect Rust code generated when run with --explicit-padding.

// bindings.h
struct repro {
    short a;
    char bits: 1;
};

First, we look at the output from bindgen without the --explicit-padding flag:

$ bindgen bindings.h

...
#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct repro {
    pub a: ::std::os::raw::c_short,
    pub _bitfield_align_1: [u8; 0],
    pub _bitfield_1: __BindgenBitfieldUnit<[u8; 1usize]>,
    pub __bindgen_padding_0: u8,
}
...

Now, we repeat it with --explicit-padding:

$ bindgen --explicit-padding bindings.h

...
#[repr(C)]
#[derive(Debug, Copy, Clone)]
pub struct repro {
    pub a: ::std::os::raw::c_short,
    pub _bitfield_align_1: [u8; 0],
    pub _bitfield_1: __BindgenBitfieldUnit<[u8; 1usize]>,
    pub __bindgen_padding_0: u8,
    pub __bindgen_padding_1: u8,
}
...

Notice that the second invocation has an extra padding field.

[sw17ch]

r? @emilio

I've seen your name around other things with bitfields, so hopefully this is familiar territory for you.