v0.8.0

Minor Changes

• 4d352a8: Add fastCapture, an opt-in that narrows the screenshot renderer's
  computed-style enumeration to a curated allow-list instead of every property
  the browser exposes. That enumeration is ~91% of a capture's cost and scales
  with element count; the list cuts the dominant phase by about 2.7x, measured
  pixel-identical to a full capture on the pages it was verified against.

  Off by default: a property the list does not name is absent from the image,
  so the trade belongs to the host. See the README for when to turn it on.

• 4d352a8: Add skipIframeContent, an opt-in that renders embedded documents as blank
  instead of cloning them. A same-origin iframe's cost is invisible from the
  host page — the renderer clones the whole embedded document, so a page of
  242 elements can be a capture of 9 245.

  The <iframe> element is kept, so its box and the layout below it stay
  where the page put them. Off by default: on a same-origin frame the content
  you lose is real.

• 4d352a8: Add captureTimeout, bounding how long one remote asset may hold a capture
  up while the renderer re-fetches the page's images and fonts to inline them.

  The default is unchanged. A dead asset URL stalls a capture for about a
  minute — the renderer's own 30 second deadline, paid twice because the number
  drives two waits in sequence, one for the image on the page to load and one
  for the fetch that inlines it — and the capture still succeeds with that asset
  replaced by a placeholder. The wait is bounded rather than multiplied: one
  dead asset and ten cost the same.

  Left at the default because a shorter one drops assets that were only slow
  and leaves holes in the image with nothing to say so. Set it if you have
  measured your own page. Only a finite positive number is honoured: 0 means
  "never give up" to the renderer, and Infinity is coerced to no wait at all.

• 4d352a8: Dragging a region no longer waits for the screenshot. The marker and the
  comment box appear as soon as the mouse is released, and the crop drops into
  a "Capturing…" slot in the attachment strip when its render lands. On a
  heavy page that removes over a second between the gesture and being able to
  type; Send waits for the crop if you get there first.

  Adds capturingScreenshot to both locales.

Patch Changes

• 4d352a8: Fix captures coming back blank on very long pages. Past the browser's canvas
  ceiling — 65 535px in a dimension, and an area cap besides — a canvas accepts
  its size, hands out a context, takes every draw call and holds no pixels, so
  every screenshot off it was empty with nothing said about it.

  Renders now fit their scale to what the browser will actually paint, verify
  that the result holds pixels, and retry smaller if it does not. Pages below
  the ceiling are unchanged; past it the capture goes soft rather than blank,
  and a render that cannot be produced at all reports through onError instead
  of attaching an empty image.

• 4d352a8: Screenshot capture no longer freezes the page. The clone traversal now hands
  the main thread back to the browser on an 8 ms budget, so the page keeps
  painting and accepting input while a render is in flight — on heavy pages
  that render could block for over a second. Wall-clock capture time is
  unchanged; what changes is that it is no longer a freeze.