fix(openai-sdk): don't inject empty system prompt when no memories are found

[devteamaegis]

Bug

In packages/openai-sdk-python/src/supermemory_openai/middleware.py, add_system_prompt() builds a memories string and then unconditionally modifies the message list — even when memories is empty (i.e. the profile and search results both returned nothing).

What happens today when no memories are found:

1. Existing system prompt present — the content becomes "You are helpful. \n " (trailing whitespace appended).
2. No system prompt present — a new {"role": "system", "content": ""} message is prepended to the list.

Both outcomes corrupt the conversation: the first wastes tokens with whitespace junk; the second injects a blank system turn that some models treat as an explicit instruction.

Minimal reproduction:

# memories resolves to "" when profile/search return nothing
messages = [{"role": "user", "content": "Hello"}]
result = await add_system_prompt(messages, ...)
# Before fix: [{"role": "system", "content": ""}, {"role": "user", "content": "Hello"}]
# After fix:  [{"role": "user", "content": "Hello"}]   (unchanged)

Fix

Add an early return before the modification block:

if not memories:
    return messages

One line. No behaviour change when memories are present.

Tests

Added test_empty_memories_do_not_modify_messages to tests/test_middleware.py covering both cases (messages with and without a pre-existing system prompt).