fix(benchmark): handle non-object baseline as CLI validation error

[97Three]

问题摘要

reposteward benchmark run --baseline raises an uncaught TypeError when the baseline file contains valid JSON whose top-level value is not an object. The invalid baseline is not handled as a normal CLI input error.

最小复现

1. Create a baseline file containing a JSON array:

tmp_dir="$(mktemp -d)"
printf '[]\n' > "$tmp_dir/baseline.json"

2. Run:

uv run reposteward benchmark run \
  --scenario context.utf8_estimate \
  --baseline "$tmp_dir/baseline.json"

3. Observe an unhandled traceback ending with:

TypeError: baseline benchmark report must be an object

预期行为

The CLI should report that the baseline must be a JSON object, return exit code 2, and not display a Python traceback.

实际行为

The exception is not handled, and the command displays a traceback ending with:
TypeError: baseline benchmark report must be an object

环境

RepoSteward commit: a6dc8c6
OS: Ubuntu 24.04.4 LTS on WSL2
Kernel: Linux 6.18.33.2-microsoft-standard-WSL2 x86_64
Python: 3.12.3
uv: 0.12.10
Harness: Codex

补充证据

load_benchmark_report() raises TypeError when the decoded JSON value is not a dictionary. The CLI's top-level error handler handles ValueError and several other expected errors, but not TypeError, so this exception escapes the normal CLI error path.

提交前确认

• 我已搜索已有 Issue 和 PR。
• 这不是需要私下报告的安全问题。
• 我已经删除凭据、账号缓存和个人信息。