[WIP] Improve parser-class cache threshold defaults

[suleman-uzair]

This PR updates adaptive cache threshold handling so parser-class grammars use a separate fallback from direct atom-level parsing.

This PR:

• adds a :parser_default threshold for parser-class contexts without a named override
• sets that parser-class fallback to 0 so memoization starts immediately for unnamed parser classes
• keeps direct atom-level parsing on the existing 1000 threshold
• keeps named parser thresholds, such as JsonParser, ahead of the parser-class fallback
• makes the fallback check explicit so 0 is treated as an intentional threshold value
• adds focused specs for atom-level defaults, parser-class defaults, and named parser precedence
• adds a benchmark-framework case that compares the new parser-class fallback with the previous 1000 behavior

This matters for recursive or retry-heavy parser-class grammars where the same atom can be attempted again at the same source position before the input reaches 1000 bytes.

This change affects the pure-Ruby parsing path only. Native mode still goes through the Rust-backed parser path and is not changed here.