Function funify_json 

pub fn funify_json(value: Value, f: &Funifier) -> Value

Recursively walk a serde_json::Value and funify every value whose containing key is NOT in Funifier::is_metric_passthrough. Returns the funified value — input is consumed (cheaper than cloning a deep tree).

Inverted polarity (metric allowlist): the default action is “funify it” — a value passes through unchanged ONLY when its containing key is a metric (count/rate/ratio/byte/duration/ structural enum). Any other field — pid, comm, cgroup_path, scheduler name, version string, novel identifier-shaped key the schema didn’t have last week — gets replaced.

Funification rules at the leaves:

• String under a non-metric key — replaced via Funifier::petname_for using the key name itself as the namespace. Two distinct keys with the same string value get different fun names; the same key + same value yields the same fun name everywhere in the dump (cross-reference preservation).
• Integer (u64 or i64) under a non-metric key — replaced via Funifier::numeric_id / Funifier::numeric_id_i64 with the key name as namespace. Sentinel zero and u64::MAX pass through unchanged (Funifier::is_sentinel_u64); the i64 path also preserves zero per Funifier::numeric_id_i64.
• Float — always passes through. Floats are quasi- exclusively rates/ratios/durations in the dump schemas (cpu_time_fraction, wakeups_per_sec, …) and there is no sensible fun mapping for IEEE-754 values; making the rule uniform avoids hazarding the rate/ratio metrics that happen to live under non-metric-keyed parents (e.g. inside an anonymous-object array element).
• Bool / null — always pass through.

Recursive rules:

• Object — re-classify each key independently. Nested objects do NOT inherit metric state across the boundary.
• Array — children inherit the parent key’s metric/non-metric verdict and (when non-metric) the parent key’s namespace. So "pids": [1, 2, 3] funifies each int under namespace “pids” and "counters": [...] passes every element through.