Optuna re-suggests identical parameter combinations and marks them as FAIL

## Problem

Optuna can suggest the same parameter set as a previous trial (including successful ones). Today we only skip failed/pruned duplicates, and we mark them as `FAIL` — wasting GPU time and polluting the study.

## Expected behavior

- Skip exact duplicates already present in the study (configurable via `optimization.no_repeat`, default `true`).
- Record skips as **PRUNED** (`skip_reason: duplicate_parameters`), not **FAIL**.
- Stop early when the search space is exhausted (grid / small discrete spaces).