Drop the causal trace's arrowheads, keep one only on the stub

[xmap]

A marker's size scales with the edge's own stroke width, so the heaviest hop
drew a head at 4.5 * 2.1 = 9.5px, wider than the 5-7px mark it pointed at:
fattest exactly where edges crowd hardest, undoing the thinning that lets the
near story read before the far one. The time axis already states direction,
since a cause is always earlier than its effect, so the head was answering a
question nothing needed answered.

Hue now carries the whole split the head used to gesture at, "why did this
happen" against "what did it set off", in both the edges and the tooltip
card's rows. The stub keeps its head: it hangs into empty space with no
second mark to orient against, so without one it would read as an error bar
rather than a pointer at an off-window cause.

Also, in the same pass:

• edgePath no longer stops short of the target mark to leave room for a
  head. Paths now run centre to centre and tuck under the marks, which the
  edge layer's paint order already allows for free.
• the pinned tooltip card is the one card a reader can scroll into now;
  pointer-events toggles off for the hovering and released states so a
  following or hidden card never swallows pointer events over the chart.
• the preview fixture moves from picking random stream/event pairs to a
  scripted "one shift at 2-BM" narrative, timed against the same snapshot
  numbers the tables already show, so a causal chain reads as a real
  sequence rather than nonsense when eyeballing this change.

No build step, lint, or test runner covers this directory (see the harness
comment); verified by serving infra/status-relay/design/page-preview.html
locally and eyeballing the scripted shift.

🤖 Generated with Claude Code

[github-actions]

Coverage report

This PR does not seem to contain any modification to coverable code.