Refactor low-level parsing and hashing

[camillescott]

• Add a kmer iterator compliant with STL InputIterator
• Add a kmer iterator that skips invalid positions
• Add a parser that does not validate on initial sequence read
• Maybe: once-again remove exceptions from low level validation, in favor of std::variant with error codes.
• Clean up and centralize exception handling
• Remove all raw C-strings from hashshifter and extenders

For the latter: exceptions in the validation and hashing code introduces approximately 20% overhead! I removed them at one point, but added them back in because of some edge cases.