Adopt descriptive task names and streamline task contributions

[hanhainebula]

Summary

Organize Search-SWE around descriptive task names and a direct contribution workflow. Task packages, evaluation tooling, asset references, and contributor guidance now follow the tasks/<task-name> structure.

• Rename all nine task packages using descriptive names of at most five hyphen-separated words. Update canonical task identifiers, evaluation references, and Hugging Face dataset paths and revision pins.
• Describe tasks through their objectives, inputs, metrics, and resource constraints. Rewrite agent-facing instructions as practical user requests, improve command formatting, and correct inconsistent task descriptions.
• Let contributors create packages directly under tasks/<task-name>. Remove the task-submissions directory and promotion scripts.
• Update authoring and maintainer skills to cover task creation, review, asset publication, and finalization. Include coding-model trials, trajectory and score analysis, feedback-driven task refinement, and revalidation. Prompt contributors to configure missing API keys locally.
• Introduce check_tasks.py with review-stage and merge-ready validation, including task naming, author attribution, package structure, and official asset-source checks. Update CI and regression tests accordingly.
• Align the English and Chinese READMEs, contribution guides, evaluation documentation, and PR template with the current workflow.

Usage changes

• Select tasks with --task <task-name>, for example --task reasoning-query-rewriting. Numeric task IDs are no longer accepted.
• Use REASONING_QUERY_REWRITING_OPENROUTER_API_KEY in place of TASK_1_1_OPENROUTER_API_KEY.
• Contribute new packages at tasks/<task-name> and publish official dataset assets under the matching Hugging Face namespace before merging.

Validation evidence

• python scripts/check_release.py: passed with 0 errors and 0 warnings.
• python scripts/check_tasks.py --merge-ready --base <main-commit-sha>: passed with 0 errors.
• Repository unit tests: 169 passed; 2 skipped because harbor-rewardkit==0.2.0 was unavailable.
• Authoring and maintainer skill regression tests: 14 passed.
• git diff main...jianlv-dev --check: passed.

Full container builds, GPU execution, and paid coding-model evaluations were not rerun for this repository reorganization.