Add agent --skill flag

[SeanAlexanderHarris]

What is the problem this would solve and/or feature it would enhance?

Providing a clean, idiomatic way for agent skill uses to pass a the skill name used for a CLI command invocation would help improve monitoring and insight into agent use of the CLI. eg. Are certain skills, Prolific or otherwise, being used for certain CLI commands to create value? Are some skills unsuccessfully using CLI commands and the API endpoints they cover?

What you are proposing to solve the problem and/or enhance the feature?

• A --skill CLI flag only (no env var, no viper binding).
• Register it once, gloabally, on the root command — so its available before every subcommand
• Include on a X-Prolific-Agent-Skill header
• The skill header should be sent on every outgoing request, including the three paths that currently build their own *http.Request and bypass the shared Client.Execute() choke point (dataset upload, collection export download, batch export download).

[SeanAlexanderHarris]

Update: the original plan above described sending the skill value via a dedicated `X-Prolific-Agent-Skill` header. That approach has been superseded — it's now folded into the standard `User-Agent` header instead (as a `skill/` token, alongside the `agent/` token added in #462), since a custom header is likely to be scrubbed by the backend before it reaches application code. See #463 for the implementation.