Improve performance by removing SHA

The performance could be faster.

When I profile my blog build, about 60% of the time spent in markdown2.py is spent running SHA hashes.

Could you explain the logic of _hash_html_block_sub and _hash_text generally? Why are we even running SHA inside a markdown converter?

It looks like... this is some some of escape mechanism, maybe? Like we generate a key, replace the HTML with the key (so it doesn't look like HTML to some other stage of the parser that should ignore it), do some processing on the outer HTML, and finally replace all the keys with the original HTML?

That could be served just as well by generating a random string rather than a hash, if so?

Ex.

def _hash_text(s: str) -> str:
    'md5-' + sha256(SECRET_SALT + s.encode("utf-8")).hexdigest()[32:]

could be replaced by the much faster

hex_digits = "0123456789abcdef"
def _hash_text(s: str) -> str:
    'md5-' + ''.join(random.choice(hex_digits) for _ in range(32)) 

for a quick fix.

(Estimate says this will make markdown conversion 2.5X faster)

Start in markdown2.py by reading _hash_html_block_sub and _hash_text and tracing how their generated keys protect and restore content during conversion. Profile the current hashing cost, then verify that any replacement preserves conversion behavior and avoids key collisions while improving performance.