Duplicate object keys resolve differently in `__getitem__` vs `to_python()`

[whyvineet]

For an object with duplicate keys, db[key] and db.to_python()[key] silently disagree with each other on the same file; one returns the first occurrence, the other the last.

Root cause

IndexStorage._child_by_key_cached (storage.py) scans children forward and returns the first match:

for i in range(start, end):
    ...
    if unpacked[2] == key_id:
        return self._make_record(child_id, unpacked, parent_offset)

to_python() (lazy.py) instead re-parses the raw bytes with json.loads, which is typical of JSON implementations, and means it keeps the last occurrence of a duplicate key.

Reproduction

import bytejson

p = "./tmp/dup.json"
open(p, "w").write('{"a":1,"a":2,"a":3}')
db = bytejson.open(p)
print(db["a"])             # 1
print(db.to_python()["a"]) # 3

Observed

db['a'] = 1
db.to_python()['a'] = 3

Expected: both access paths should agree (standard practice is "last key wins", matching json.loads and to_python()).