Also benchmark mismatches

[almondtools]

Currently the naive algorithm seems to perform quite well for large alphabets and small patterns. There are some reasons for that:

• small patterns means that backtracking cost is low
• most patterns to be searched can be found in the text
• large alphabets lead to early backtracking (most break at the first or second character)

To get a more balanced benchmark, we should modify it by

• inserting mismatching patterns
• inserting patterns that mismatch at the last pattern position