mitmweb: make flow table columns resizable - #8358

Fixed table layout hands the space that sized columns leave over to every column without a width of its own, so setting one width while the others stayed flexible made the browser inflate all of them: columns set to 100px rendered at 195px and the handle tore away from the cursor.
Every column is now pinned when a drag starts and quickactions absorbs the slack, so the width follows the pointer exactly, and widening past the viewport scrolls horizontally instead of being clipped.
Pointer moves are coalesced into one update per animation frame and localStorage is written once at the end of the drag rather than on every event.
Adds pointer capture, a pointercancel teardown, and a body class that blocks text selection while dragging.

The handle of the last column sat flush against the right edge of the table, where the scrollbar and the pane splitter leave nothing to grab, and the column that takes the width the others leave over carried no header at all, so it read as a stray divider beside an unlabelled column.
That column is now headed Actions after what its cells hold, and it keeps absorbing the slack, which is why it is neither sortable nor resizable.
A narrow filler column keeps the last handle clear of the scrollbar for as long as the actions column is still collapsed, and the rows carry a cell for it because a column without cells is not covered by the row background.
Widths may only sit on the col elements: fixed layout falls back to the first row's cell whenever a col is auto, which would pin the one column that has to stay flexible.

The last onResize and onResizeEnd both ran synchronously from the pointerup listener, where React batches state updates, so the width that got stored and the layout that got re-measured were still the ones from the previous animation frame.
A drag that finished inside a single frame left nothing behind at all.
The end callback now carries the final width, and the table stores and re-measures it from the setState completion callback.

jsdom does not implement PointerEvent, so fireEvent.pointerDown and friends fall back to a bare Event and every clientX the test passed was dropped.
The drag arithmetic ran on undefined throughout and the test could only assert which keys came back, never a width.
Dispatching MouseEvents under the pointer event types carries the coordinates through, and stubbing offsetWidth gives the columns a width to start from, so the reported pixels can be asserted.

The 20px floor a drag is clamped to sits above the 10px the TLS column ships at, so pressing and releasing on its handle without moving doubled its width.
Any restored width below the floor jumped the same way.
A column narrower than the minimum now keeps its own width as the floor, and a gesture that never moved the pointer reports no new width at all.

Comparing the last pointer position against the starting one treated a drag that happened to end on the pixel it began from as a gesture that never moved, so the pending frame was cancelled and no width was reported.
The column stayed at whatever width the last frame before that had applied, and that width was the one persisted.
An explicit flag set on the first pointermove answers the question the comparison was standing in for.

The actions column ships collapsed until it is the one taking the slack, but its header carries the Actions label all the same, and a th never clipped its content the way the body cells do.
The label rendered past the right edge of the table, so the default state, before any width has been dragged, opened with a horizontal scrollbar over 28px of overflow.
Clipping the header content keeps every label inside its own column, which also covers a column dragged narrower than the word it is headed by.

The row click handler bailed out of selection for any click inside the quickactions cell, not just on its buttons.
That was invisible while the column was collapsed to the width of its buttons, but a resized column leaves empty space that stopped selecting, and rows with no quickaction at all were unselectable across the column.
Guard on the buttons themselves instead.

The width state, its localStorage round trip and the resize handle's click guard had no test, so those lines showed up as a patch coverage drop.

A drag had a floor but no ceiling, so a column could be pulled well past the right edge and the layout it left behind was hard to drag back.
A column that is already wider, restored from a wider window, keeps its width as the ceiling and may only shrink.

There was no way back to the default layout once columns had been dragged.
The widths move from FlowTable's own state into a ui/columnWidths slice so the Flow List menu can reach them, which also puts the localStorage round trip in one place.