max / quasi

0.69.2: a renderer says where it drew its rows `600c9e42` asked what, if anything, a description should hand a host about where it drew what. Nothing, and it is the same rule as the rest of the day's: `Act::shows` carries a picture and not a box, and `ae8e8836`'s three candidate anchors all name a described thing rather than a pixel. A description says what is on the screen and never where. Geometry is the renderer's answer and a different answer per host -- a terminal has cells, a webview has the DOM, egui has rects -- so the fact travels from the renderer to its own host beside the drawing rather than through the description. A webview host asks the DOM the same question and needs nothing from here. That constraint is shared with `ae8e8836` and does not pre-empt its choice between (a), (b) and (c), all three of which are compatible with it. `quasi_immediate::row_at(ctx, pos) -> Option<RowAt>`, carrying the row's value (`None` where the description gave none) and its index. Every described row is noted whether or not it answers a gesture: "is the pointer over a row" has to be answerable for the row that answers nothing, which is the dangerous case. Through egui's own context data rather than a return value. `Immediate::screen` and `chromed` answer `Option<Fired>`, which is what the user set off; widening that would make every call site in every host thread a fact almost none of them want. Per-frame data belonging to a context is what context data is for. Rects are keyed by the pass they were measured in and cleared when it moves, since this crate has no begin-frame hook. A table cell is noted after it has drawn, because a `Ui`'s `min_rect` is empty until something is in it and measuring first recorded a sliver at the cell's left edge -- caught by the test rather than reasoned about. `table_columns` split out of `table` while it was over clippy's line bound. It was already a self-contained fact about the two vocabularies meeting: a tick takes no column in the description and this renderer adds one.