Fixes T9268. Currently, we try to match any string like "a2f313f1" as a commit/revision, so short hashes will get picked up.
However, we don't require a word boundary or terminal after the match, so for input like "aaa...aaaaz" the engine can get stuck trying to split the string into sub-matches.
That is, in the original case, the input "aaaz" had valid matches against [rA-Z0-9a-f]+ up to "z" of:
aaa aa a a aa a a a
All of these will fail once it hits "z", but it has to try them all. This complexity is explosive with longer strings.
Instead, require a word boundary or EOL after the match, so this is the only valid match:
aaa
Then the engine sees the "z", says "nope, no match" and doesn't have to backtrack across all possible combinations.