bench: add stats, token-count, and safe-div benchmark cases

[Aditya251610]

Adds three small agent-oriented benchmarks from #8:

• stats — min/max/sum over a fixed buffer (loops, arrays, comparisons)
• token-count — delimiter/token scan over a string span (byte scanning)
• safe-div — typed error path with rescue on division by zero (fallibility)

Each follows the existing benchmark pattern: tight loop, deterministic result, expected
stdout.

Addresses #8

[vercel]

@Aditya251610 is attempting to deploy a commit to the Vercel Labs Team on Vercel.

A member of the Team first needs to authorize it.

[Aditya251610]

Now this should close the issue #8