README: add leaderboard section, drop per-task CWE table
Adds a Leaderboard section with the current results on the 100 public tasks (11 models) and a link to the CWE-bench leaderboard Space (https://huggingface.co/spaces/collinear-ai/cwe-bench-leaderboard).
Why a Space instead of the Hub's built-in benchmark leaderboard: the built-in leaderboard only aggregates .eval_results/ files from each model's own repo. Most CWE-bench models are API-only (Gemini, Claude, GPT, Grok, Qwen Max) and have no model repo, so they can never appear there; the dataset's built-in leaderboard currently has 0 entries. A static Space lets us show every evaluated model and update it by editing one results file.
Also removes the per-task CWE coverage table, and drops "difficulty" from the task.toml description (that field is removed in #2).
