fix normalize very long sequences

[bnika]

The normalizer now handles very long sequences like this: C4CE6358DF86040CAAEEBC58951B8E133105D3369980D62D109E5B6B75CCE67C_713x0

• don't limit "separate uppercase letters" to 5 letters like previously (just use '+' in the regex instead of {1,5}
• let the tokenizer separate on '_', remove the underscore when normalizing alphabet at the end
• lowercase everything in the end

The last change influences all normalization since the toLowerCase() is now called as last operation before returning the normalized string