add BOTAN_NORETURN macro and mark those functions that do not return with it

[neusdan]

#402

[randombit]

The build.h modification looks fine as does the assert.

I do not think noreturn is appropriate on any of the encode_into calls; they are throwing simply due to encoding of those values not being implemented, but I don't want to commit to them not ever being implemented in the future with a noreturn API contract.

Similarly it does not seem right on Credentials_Manager::psk, since while the default throws, a subclass might well return if a PSK is defined. I have no idea what a compiler does with a virtual noreturn, but I would guess it either ignores the noreturn or happily generates code assuming no subclass will ever return either. I find it kind of weird GCC would suggest this one.

For http_transact_fail it makes sense, but I wonder why on earth this is exposed to callers? It is just an internal implementation detail I'm pretty sure. (I did this, I just can't remember what I was thinking at the time.) Probably it should be pulled out of the header entirely and be an anon namespace function in the source file, as there is no reason for any application to ever call it. I'll handle this.

Please trim the PR to just buildh.in and assert.h for now.

[neusdan]

Good points. What about the warning flag? Do you wan't to have it enabled?

[neusdan]

somethings wrong with the amalgamation build

[neusdan]

ok, that was dumb ;-) it goes before the function declaration