feat(embeddings): support OpenAI-compatible embeddings providers

[yordis]

• Operators want to point Memorizer at hosted OpenAI or any OpenAI-compatible gateway (LiteLLM, vLLM, Azure OpenAI, LocalAI) without standing up Ollama, so embedding generation matches the rest of the LLM ecosystem they already operate.
• The bifurcated provider model already exists for the Memorizer Agent side; aligning the embeddings side keeps configuration shape consistent and unblocks credential-bearing providers via an optional ApiKey.
• Drop-in for existing Ollama deployments: Provider defaults to ollama, no schema migration, no required new settings, no change to the Ollama wire format.

[yordis]

[yordis]

[yordis]

docker exec memorizer-postgres-dev psql -U postgres -d postgmem -c \
  "SELECT id, title,
          vector_dims(embedding) AS dims,
          substring(embedding::text from 1 for 120) || '...' AS preview
     FROM memories
    WHERE id = '3ede201f-4a1f-4c6a-bc29-de3ae1b752ef';"

Result
                  id                  |         title         | dims |                                                           preview
--------------------------------------+-----------------------+------+-----------------------------------------------------------------------------------------------------------------------------
 3ede201f-4a1f-4c6a-bc29-de3ae1b752ef | OpenAI MCP smoke test | 1536 | [0.0435791,-0.02798462,0.05859375,-0.024337769,-0.05697632,-0.025756836,0.0446167,0.01134491,0.0065956116,-0.01576233,0....
(1 row)

[Aaronontheweb]

Opened replacement PR #194 on top of this work: https://github.com/petabridge/memorizer/pull/194\n\nAdditional changes included there:\n- Preserve the OpenAI-compatible embedding provider implementation from this PR.\n- Apply active embedding provider settings before dimension validation on save/activation so migration checks use the selected provider shape.\n- Keep embedding API keys sourced from runtime configuration only; provider_settings JSON no longer stores API keys.\n- Redact provider config responses and recursively strip sensitive keys before provider config is stored.\n- Add a migration to remove any previously persisted provider API keys from provider_settings.config.\n- Scope bearer auth to individual OpenAI-compatible embedding requests so keys are not sent to Ollama endpoints.\n- Make fallback embedding dimensions schema-safe during dimension migrations.\n- Add focused tests for request auth behavior and provider config sanitization.\n\n#194 includes Closes #193 so this PR should close automatically when the replacement merges.

[Aaronontheweb]

Closing in favor of the merged replacement PR #194, which preserves this contribution and adds the provider migration / credential handling hardening needed before landing.\n\nMerged replacement: #194