Avoid introducing typecasts from C `char` to Rust `libc::c_int` in comparisons to literals

[64kramsystem]

In the source codebase I'm porting, there are char variables, that are compared to char literals.

In the transpiled codebase, the compared values are (explicitly) casted to i32, which is redundant.

Now, I can imagine the reasons why this happen - if C2rust works on some IR, on a modern architecture, comparing a 32 bit value may be preferrable to an 8 bit one, however, in the context of transpiling a project, faithfulness is paramount.

Reference code follows.

Original:

// function prototype
int get (void);

// Global
char ch;

// Code of interest
do {
  ch = get();
} while (ch<'0' || ch>'9');

Translation (slighly simplified):

// Global
static mut ch: libc::c_char;

// function signature
fn get() -> libc::c_int;

// Code of interest
loop {
    ch = get() as libc::c_char;
    if !((ch as libc::c_int) < '0' as i32 || ch as libc::c_int > '9' as i32) {
        break;
    }
}

Ideal code of interest

loop {
    ch = get() as libc::c_char;
    if !(ch < '0' as i8 || ch > '9' as i8) {
        break;
    }
}

[kkysen]

I believe the reason this is done is because character literals in C, like '0', have type int. Thus, '0' in C is transpiled to '0' as i32 (I'm not sure why that's not '0' as libc::c_int). Then to do the comparison, we also cast the LHS to libc::c_int (sign extension of char widening in C is implementation-defined, and we just use the Rust as implementation here).

[64kramsystem]

Right!

[kkysen]

I do think ch < '0' as libc::c_int would be better and equivalent in this case, but that might be much easier to refactor outside of the transpiler.