librustls: switch common.c's read_file to binary mode

[cpu]

Prior to ECH support all of the consumers of the common read_file() helper were loading text data (PEM encoded certs, keys, and CRLs). However, for ECH, we're loading a binary serialized ECH Config List structure.

On Windows specifically if we don't fopen() with mode==rb instead of mode==r we open the file in text mode, and certain byte sequences are translated:

In text mode, carriage return-line feed (CRLF) combinations are translated into single line feed (LF) characters on input,
...
When a Unicode stream-I/O function operates in text mode (the default), the source or destination stream is assumed to be a sequence of multibyte characters. Therefore, the Unicode stream-input functions convert multibyte characters to wide characters (as if by a call to the mbtowc function).

This in turn breaks ECH config deserialization when the config data happens to have those byte sequences. This manifests as a Windows-only flake of the ECH example unit test in CI, with the error:

client[8988]: failed to configure ECH with any provided config files

The fix is simple: we can use 'rb' as the mode in read_file() and avoid this issue.

Resolves #514

[cpu]

rustls-ffi / Clippy nightly (optional) (pull_request) Failing after 1m
rustls-ffi / Check for documentation errors (pull_request) Failing after 1m

Both need #649 to land first

[cpu]

Not that it demonstrates too much except a lack of regressions, but here's a manually invoked daily tasks job passing on this branch.