fix(memory): skip malformed encrypted session envelopes

[hsusul]

Summary

EncryptedSession documents that invalid or expired stored envelopes are skipped, but malformed payload types, invalid UTF-8, invalid JSON, and valid JSON with a non-object shape could abort session reads. Harden _unwrap to skip those malformed envelopes while preserving valid response-item mappings.

Test plan

• Added regression coverage for non-string, lone-surrogate, and non-object encrypted payloads.
• Focused: 21 passed (tests/extensions/memory/test_encrypt_session.py).
• Full verification: UV_CACHE_DIR=/tmp/uv-verify-pr5-a2 UV_DEFAULT_INDEX=https://pypi.org/simple bash .agents/skills/code-change-verification/scripts/run.sh — all commands passed (format, lint, typecheck, tests).

Issue number

N/A

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4beb5f5bf4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Handle ValueError from oversized JSON integers

When an authenticated envelope decrypts to a JSON object containing an integer longer than Python's configured digit limit (for example, more than 4,300 digits by default), json.loads() raises ValueError, not JSONDecodeError. That exception escapes here, so get_items() and pop_item() still abort instead of isolating the malformed record as intended; catch this deserialization failure at the same boundary.

AGENTS.md reference: AGENTS.md:L165-L166

Useful? React with 👍 / 👎.

[seratch]

Thanks for investigating this. Normal expiry, key mismatch, and ciphertext corruption already fail Fernet authentication and are skipped through the existing InvalidToken handling. The new branches require a structurally corrupted typed envelope or plaintext that was validly encrypted with the session key but was not produced by EncryptedSession._wrap().

I am going to close this PR for now because no supported backend, migration, or user report has been shown to produce those shapes. If you can provide a concrete stored record from a supported session backend and trace how ordinary SDK operation produced it, we can reconsider the narrow corruption boundary that needs handling.