Require end-of-input to be reached for range-based parser invocations

[dominiklohmann]

Motivation: We use a custom-made parser-combinator framework for high-performance parsers in VAST, and to combine parsers one needs to be able to resume parsing after partial consumption of input. Experience has shown (e.g., #791) that we don't usually want to silently discard input on the resulting, combined parsers.

Before the change implemented by this PR, EOI checks were required to be done manually by the caller, e.g. by calling (p >> eoi)(rng, attr) over p(rng, attr). Outside of combining with the eoi parser, there was no way to verify whether the EOI was reached successfully for range-based parser invocations. This has turned out to be very error prone.

For [begin, end)-iterator based invocations, nothing changes. It is easy for the caller to verify whether the EOI was reached successfully by simply comparing begin and end.

Among other things, this PR completely disables the MRT integration, which we knew to be broken, but which silently succeeded in parsing invalid data. It will be revisited at a later point in time.

[dominiklohmann]

End result:

• CSV map parser is broken; test has been disabled.
• MRT integration is broken; has been removed (for now). This was known to be broken, it just never failed in the tests.
• A whole bunch of parsers were fixed by the second commit, which strips the NUL byte from parser invocations using a known-length char array. This mostly affected our test suite.

[tobim]

Looks good, I'm surprised there are not more breaking tests.