M3: Eval task format and the first ten tasks

A tasks/ directory where each task is a directory containing a prompt, a pytest file, and a manifest declaring the expected entry point.

Ten tasks minimum, deliberately spanning the range from trivially easy to reliably-failing, so the suite has headroom in both directions. A suite where everything passes and a suite where everything fails are equally uninformative. Draw from real intro-Python coursework: string formatting, loops and conditionals, a small class, file I/O, a recursive function, a function with awkward edge cases.

Done when the tasks exist with their tests, and each test file has been verified against a known-good reference solution.

Start by inspecting the repository and the proposed tasks/ directory structure, then identify how pytest files and manifests should describe each task's expected entry point. Create at least ten introductory Python tasks spanning the listed difficulty range, with prompts, tests, and manifests. Verify every test file against a known-good reference solution and confirm the suite includes both passing and reliably failing cases.