Remove double LLM call in `--simple` mode to reduce cost and latency

[calchiwo]

The current --simple mode flow:

1. Generate full explanation via LLM

2. Build a new prompt from that output

3. Call the LLM again to summarize

This creates two problems:

1. Doubles latency

2. Doubles token usage and cost

As repository size grows, the first output may become large. Feeding that entire output back into the model increases:

• Token consumption

• Risk of hitting context limits

• Hallucination amplification

This is unnecessary for simple mode.

--simple should not depend on generating the full explanation first.

Proposed change:

When --simple is set, build a simplified prompt directly from:

• Repo metadata

• README

• Optional tree summary

Then call the LLM once.

Benefits:

• Single LLM call

• Lower cost

• Lower latency

• Reduced token overflow risk

• Cleaner execution path

This keeps --simple lightweight, which aligns with user expectations.

Right now it behaves like “generate detailed, then compress,” which is inefficient by design.