Feature: Improve efficiency of T3Code Session title generation

[alastairlundy]

For each new thread in T3Code, title generation is performed by the model specified in the settings menu. Title generation in T3Code works by creating a new session in the relevant harness to generate the title, however this leads to additional unintended AI spend. If the underlying harness T3Code wraps uses its own title generation mechanism for the session T3Code creates (such as OpenCode), then extra AI usage is spent by the harness being wrapped on giving a session title to a session nobody will ever see or need besides T3Code (the title generation session).

Even though OpenCode now uses GPT-6 Luna for their session title generation, having T3Code spawn a session in OpenCode for title generation wastes unnecessary cost during title gen to name the title gen OpenCode session.

For 1 session the cost is negligible (partificularly if using OpenCode with GPT-6 Luna) but over time the cost can add up.

I suggest T3Code internalize title generation behaviour, rather than relying on wrapped agent harnesses for this, to avoid this unnecessary and arguably unintended AI spend during title generation.

Example:

1. New thread in T3Code started, harness being wrapped starts work
2. T3Code spawns title gen session in wrapped harness
3. Wrapped harness generates the harness title for T3Code's spawned title gen session <- Unnecessary AI spend when using T3Code
4. Wrapped harness session generates T3Code session title

[abdelrahman-aboelnour]

+1, allowing initial title to be provided by the backend ( eg. OpenCode) while optionally having the title generation seema much better.

Since in openCode alot of people (myself included) have already instructions baked into agents to name conversations in a specific way that works directly and easily from intiap prompt texr and agent choice