Data_encoding for Micheline is more permissive than tokenizer

As a follow up to issue #1930 (closed) and !3720 (merged), we should change the Micheline encoding to forbid values that for identifiers that are forbidden by the tokenizer in lib_micheline/micheline_parser.ml.

This should be as simple as adding a with_decoding_guard similar to the one introduced by !3720 (merged) for annotations. The difference is that idents only allow [a-zA-Z0-9_].