Encode per-endpoint pagination bounds in the operation declarations

Follow-up to #38.

`PageArgs::resolve` applies one global pagination policy (defaults, clamp to 1..=100), but several endpoints have their own bounds the declarations don't express:

- `GET /v2/issues` clamps `count` to a minimum of 5
- `GET /revisions/{locator}/snippets` caps `pageSize` at 50 (currently enforced ad hoc in the `Snippet` `list_page` impl)

Idea: encode per-entity bounds in the param structs / declarations themselves so that

1. both surfaces enforce the same bounds,
2. the MCP input schemas advertise them (agents see constraints instead of discovering them via API errors), and
3. `--help` makes the CLI act as documentation for what each endpoint supports.

Not needed for #38 — deliberately deferred.

Start with PageArgs::resolve and the parameter structs or operation declarations, then inspect the Snippet list_page implementation for its existing pageSize cap. Check how those declarations generate MCP input schemas and CLI --help output. Done means per-endpoint bounds are expressed once and enforced consistently across both surfaces, with the constraints visible in schemas and help.

Written by the indexing model from the issue text.