Moved the parsers into a hash table

[Witiko]

This series of commits moves parsers into a separate hash table as
discussed in #24. The average CPU time in seconds measured by
make bench (500 samples) went up from 0.617380 to 0.632300.
This is likely due to the hash table access overhead as you predicted.

[jgm]

Thanks. That seems an acceptable performance loss!

[craigbarnes]

Wouldn't it make more sense to use an LPeg grammar table for this?

[Witiko]

Lunamark already uses a grammar table (see the syntax variable in
lunamark/reader/markdown.lua). However, it contains only the top-level rules.
The users can currently supply a function that makes modifications to the
syntax table (see the alter_syntax option in lunamark/reader/markdown.lua),
which arguably makes the top-level rules a part of the library interface. By
including all the helper patterns into the syntax table, we would be exposing
much more than necessary and any changes could break user code.

The helper table could be rewritten as an LPeg grammar table, although a couple
entries would need to be removed, since the table contains not only LPeg
patterns, but also a number of functions that return LPeg patterns. However,
what do you hope this would accomplish considering that the helper patterns are
non-recursive?