Reduce uses of unwrap/expect

[josephlr]

The current code uses unwrap() and expect() more than it strictly needs to. A better approach would be to have the function return an error in these cases.

Specifically, the error type could be changed to be an enum with variants for:

• IO error, when the syscall fails
• Invalid UTF8
• Invalid Null Byte
• other error conditions we’re currently just panicking on

As this would change the error type, it would be a breaking change. However, if this was done correctly, a lot of the code where we currently use unwrap could just use “?”

[mathstuf]

FWIW, we're down to 6 unwrap and 3 expect calls outside the test suite.

Unwrap calls:

• parsing description bits as uid and gid (2 calls)
• parsing permission bits from ascii hex
• assuming utf8 for description and security descriptions (2 calls)
• construction of a CString from a &str

Expect calls:

• the kernel returning a key serial outside of the i32 range
• the kernel handing us back a key serial of 0
• a size result is outside of usize range