Fix uri transformer slowness over large buffers

[roblourens]

Fix #138784

tl;dr of that issue

• I have a notebook with a really large output, like an image that is many MBs
• The notebook is dirty
• When I save, or autosave kicks in, we transfer the entire notebook model data to the EH to serialize
• This hangs up the EH process for a long time
• I try to execute a simple cell and it takes a long time, because the EH is stuck

Turns out that most of the CPU time is in _transformIncomingURIs because it is iterating the large output VSBuffer. There will never be a URI in there, so this changes skips them.

As a separate optimization, we could also avoid transferring the full notebook model to save, since the EH should already have this model which has been updated incrementally

[jrieken]

lgtm

fyi @alexdima who wrote this