explore: cursor can leave the viewport on a one-row terminal, and ctrl+d on an empty list sets cursor to -1

Two boundary cases in cmd/openapi/internal/explore/tui/model.go, found while adding key-handler tests for #254. Both are on main today and unrelated to the bubbletea v2 change.

1. ensureCursorVisible cannot fit the cursor on a one-row content area. With height: 8, calculateContentHeight() returns 1. Once the cursor is past row 0, any non-zero scrollOffset costs one line for the "more items above" indicator plus one for the cursor row, which never fits in 1, so the search loop finds no offset and leaves scrollOffset = 0 with the cursor off-screen. Measured: height=8, cursor=5 -> scrollOffset=0, visible rows 0..0. The test TestUpdate_HalfPageKeys/ctrl+u still moves one row on a tiny terminal documents this by skipping the viewport assertion.

2. ctrl+d on an empty operation list sets cursor = -1: maxItems = len(m.operations) - 1 is -1 and the clamp assigns it. Unreachable through the CLI, since explore and snip both refuse an empty list before starting the TUI, so this is hygiene rather than a user-visible bug.

Suggested fixes: for (1), either let the cursor row win over the indicator when only one row fits, or have calculateContentHeight reserve the indicator line so the loop's budget is honest; for (2), the same len(m.operations) > 0 guard G already has. Happy to send a PR for both once #254 lands, so the tests can assert the tight bound everywhere.

Start in cmd/openapi/internal/explore/tui/model.go, focusing on ensureCursorVisible, calculateContentHeight, and the ctrl+d handling. Run the key-handler tests, including TestUpdate_HalfPageKeys/ctrl+u still moves one row on a tiny terminal; done means both boundary cases stay within valid cursor and viewport bounds and the tests assert those bounds.