fix(vault): replace hand-rolled XOR+HMAC with real AES-GCM (#302)

[aaniya22]

Description

VaultCrypto previously encrypted secrets using a hand-rolled scheme: a SHA-256-derived keystream XOR'd against the plaintext, authenticated with a separate HMAC. Because the keystream was cycled to cover plaintexts longer than one block, any secret over 32 bytes reused keystream bytes — a crib-dragging vulnerability if an attacker could influence or guess part of the plaintext.

This PR replaces that scheme with AES-256-GCM (via the cryptography library):

• VaultCrypto now uses AES-256-GCM for authenticated encryption instead of a custom XOR keystream + HMAC scheme, closing a crib-dragging vulnerability on plaintext longer than one keystream block.
• Key material is SHA-256 hashed to a fixed 32 bytes, so any input key length works safely for AES-256.
• Ciphertext from the old XOR+HMAC format is now rejected loudly on decrypt (raises ValueError) rather than silently producing garbage.
• Consolidated a duplicate test_vault.py (one had been added at testing/backend/test_vault.py, colliding with the existing file at testing/backend/unit/test_vault.py) and merged in new regression tests: long-plaintext round-trip, nonce non-determinism, tamper detection, wrong-key rejection, and old-format rejection.

Note: this is a breaking change for data at rest — secrets encrypted under the old format cannot be decrypted with the new code and must be re-encrypted as part of deploying this change.

Related Issues

Closes #302

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update

How Has This Been Tested?

Ran the full backend suite (python -m pytest testing/backend -v). All 6 tests in testing/backend/unit/test_vault.py pass, covering:

• Basic encrypt/decrypt round-trip
• Round-trip on plaintext longer than one AES block (regression test for the original vulnerability)
• Non-determinism (same plaintext encrypts to different ciphertext each time, due to fresh nonces)
• Tampered ciphertext is rejected
• Wrong key cannot decrypt
• Ciphertext in the old XOR+HMAC format is rejected rather than silently corrupted

32 unrelated failures exist elsewhere in the suite (plugin/tooling integration tests, e.g. test_phase2_plugins.py, test_phase3_plugins.py, test_plugins.py). Verified these fail identically on unmodified main, so they're pre-existing environment/tooling issues (likely missing external CLI tools) unrelated to this change.

To reproduce: python -m pytest testing/backend/unit/test_vault.py -v

Checklist

• My code follows the code style of this project.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.

[utksh1]

Moving from hand-rolled crypto to AES-GCM is the right direction, but this revision regresses existing vault behavior:\n\n1. Preserve compatibility with the current vault key contract and keep the existing fingerprint/diagnostics behavior intact.\n2. Do not remove unrelated dependencies from backend/requirements.txt (for example httpx / openai) as part of this PR.\n3. Keep the crypto migration focused and regression-safe.\n\nPlease revise this to preserve current diagnostics/contracts while upgrading the crypto implementation.

[aaniya22]

Moving from hand-rolled crypto to AES-GCM is the right direction, but this revision regresses existing vault behavior:\n\n1. Preserve compatibility with the current vault key contract and keep the existing fingerprint/diagnostics behavior intact.\n2. Do not remove unrelated dependencies from backend/requirements.txt (for example httpx / openai) as part of this PR.\n3. Keep the crypto migration focused and regression-safe.\n\nPlease revise this to preserve current diagnostics/contracts while upgrading the crypto implementation.

upstream/main already has an equivalent (and more complete) AES-GCM
implementation merged, including the key_fingerprint diagnostics property this PR
was missing.rebasing my branch onto upstream/main showed vault.py and test_vault.py now match
upstream exactly with no meaningful diff left, so there's nothing left to contribute
here.
thanks for the review

[utksh1]

Closing this because the author confirmed the AES-GCM vault change is already present on main and there is no meaningful diff left to merge.