docs: add implement-comet-expression Claude skill

[andygrove]

Which issue does this PR close?

Closes #.

Rationale for this change

We already ship two Comet-specific Claude skills under .claude/skills/: audit-comet-expression for auditing existing expressions and review-comet-pr for reviewing PRs. There is no companion skill for the most common contributor task, which is implementing a brand new expression. This PR adds one.

The skill is intentionally a thin wrapper around docs/source/contributor-guide/adding_a_new_expression.md rather than a duplicate. It encodes the workflow conventions that are not captured in the contributor guide on their own:

• Always read the latest Spark master first via a shallow clone, so the canonical behavior comes from upstream rather than from whichever Spark version the contributor happens to have checked out.
• Run the existing audit-comet-expression skill against the initial implementation to drive a structured test-coverage iteration loop.

What changes are included in this PR?

Adds .claude/skills/implement-comet-expression/SKILL.md, a short skill that:

1. Points the implementer at the relevant contributor-guide pages.
2. Walks through cloning Spark master and locating the expression class and tests.
3. Summarizes the implementation steps (Scala serde, registration, optional scalarFunctionExprToProtoWithReturnType, native scalar function, initial Comet SQL Test).
4. Hands off to audit-comet-expression for gap analysis.
5. Iterates on audit-recommended tests until the user is satisfied.
6. Finishes with the standard pre-PR make format and cargo clippy checks.

How are these changes tested?

This is a documentation-only change, so no automated tests are added. The skill was reviewed for accuracy against the current contributor guide and the behavior of the existing audit-comet-expression skill.