exporter: scrape dicts via items(), not keys() + __getitem__ (KeyError on mutated keys)

[mahmoud]

Problem

dump_graph dies with KeyError(<key>) on any heap containing a dict whose key was mutated after insertion (hash no longer matches its bucket). Iteration over such a dict works fine, but lookup by the key raises, and the exporter scraped dicts via keys() + dict.__getitem__(obj, key).

Seen in the wild via openpyxl: StyleArray is an array subclass with a content-based hash that openpyxl mutates in place after using it as a dict key. Any worker heap containing a loaded-and-modified workbook made the dump fail:

KeyError(StyleArray('i', [0, 0, 0, 0, 0, 1, 0, 0, 0]))

Minimal repro against master:

from openpyxl.styles.cell_style import StyleArray
k = StyleArray()
d = {k: 1}
k[5] = 1  # in-place mutation changes hash
from objex import dump_graph
dump_graph('/tmp/dump.db')  # KeyError(StyleArray(...))

Fix

Scrape dicts with dict.items(obj): reads key and value straight off the bucket, no re-lookup; also one C call instead of one per key.

Test

Added MutableHashKey (a pure-Python stand-in for StyleArray) to the shared fixture heap with a post-insert mutation, plus test_dump_graph_survives_mutated_dict_keys following the existing survives_* pattern. Pre-fix, the fixture dump subprocess crashes (22 errors); post-fix, the full suite passes (30 passed).