[Error handling] Put extra logging and propagate backtraces

[ivmarkov]

For a certain set of errors, specifically ErrorCodes:

• Crypto
• TLSStack
• BtpError
• MdnsError
• NoNetworkInterface (we should actually rename this to NetworkError)
• StdIoError
• maybe a few more...?

... we should:

• Always grab the original backtrace (if feature backtrace is enabled)
• ... and keep the original error as a Box<dyn Error> (if feature std is enabled, and also in future if feature alloc is enabled given that ). Note that we currently preserve the original error only if BOTH std and backtrace are enabled, which is probably a bit too paranoid
• Log the original error with error! otherwise

... to get extra context for debugging. Because these errors are:

• Wrappers for lower-level errors in their respective stacks
• We need extra info what happened for debugging

The cost of preserving the original error should not be that high as we don't expect these errors to happen frequently, but unfortunately we can only due it in the presence of alloc. Without alloc, preserving the original errors means generics which will virally pollute all our APIs which is just plain impossible to handle.

We might even decide to model these errors in a different way from the rest. To be decided.