fix(lazy): clear the key buffer before a nested get_many lookup

[yuefdev]

Fixes #241.

parse_str only clears the key buffer right before it copies an escaped key into it. A key with a simple escape (\f, \n, \) leaves its unescaped bytes in the buffer, and when get_many_keys is entered again with that buffer, debug_assert!(strbuf.is_empty()) fires. There are two ways to get there:

• a later key in the same object has no escape (so it's borrowed), matches, and recurses: {"\f":9,"c":{"d":1}} with c/d, as in the issue
• the escaped key is the last one read in an earlier array element, and a later element is looked up: {"a":[{"x":1,"\f":2},{"y":3}]} with a/0/x and a/1/y

Clearing the buffer only before the recursion in get_many_keys would miss the second case, so this clears it on entry to get_many_keys in place of the assert. Nothing reads a key's bytes after its own lookup, so the result is unchanged. Release builds were already returning the right values.

Added test_get_many_after_escaped_key covering both cases for get_many and get_many_unchecked. It panics on main with debug assertions and passes with the fix. cargo test and cargo clippy --all-targets --all-features -- -D warnings pass locally.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}