Fixes stack context for json lines format that recovers from invalid JSON lines

[elstehle]

Description

Addresses #14282.

For the JSON lines format that recovers after an invalid JSON line, we've had two issues when we were generating the stack context that is used downstream in the full JSON pushdown transducer.

For that format, we need to make sure that we "reset" the stack context after each JSON line. That is,

1. We need to reset the stack to the empty stack after each JSON line, as the stack may not be empty after an erroneous JSON line. E.g. {"this opening brace is never closed":123\n{"<=this brace should be on the empty stack":...}
2. We need to reset that we are outside of a string: {"no matching end-quote on this line\n{"<=this quote is the beginning of a field name, not the end of the previous line's field name"

This fixes above requirements as follows:

1. Was already implemented - but with an inappropriate scan operator that is not associative:

StackLevelT new_level = (symbol_to_stack_op_type(rhs.value) == stack_op_type::RESET)
                              ? 0
                              : (lhs.stack_level + rhs.stack_level);

E.g. ({,n,{,},n,{,{,n,},},n,} all fail the associativity test). This was replaced with a ScanByKey that would start with a "fresh" stack level with each new key segment.

2. Was addressed by changing the transition table of the finite-state transducer that filters out brackets and braces that are enclosed in quotes to go back to the OOS (outside-of-string) state after every newline. This behaviour requires that every newline character is treated as a delimiter of a JSON line. This was confirmed by Spark Rapids, who is the targeted user for the recovery option to be the case.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vuule]

looks good. just one question to confirm my understanding

[karthikeyann]

Looks good to me.

[elstehle]

/merge