For reading, introduce the readFileStream capability that reduces
the overhead of low level file reading by switching to an event based
model.

For writing, try to consume up to N chunks of data from the underlying
source and directly call writeFile if the source is consumed. This will avoid
the overhead of low level file reading for small to medium files.

Part of #83836

Fixes #77803

Outline "Sort By: Type" -> "Sort By: Category"

Fixes #83704

For #83703

In #81718 we adjusted breakpoint display logic to show an 'empty' arrow on the
line, and the arrow with a dot at the column where the debugger is running.
However, not all debuggers support column breakpoints. Bowden noticed, while
working on his Python demo for ignite, that breakpoints in Python code no
longer displayed correctly.

This PR tweaks the logic so that if we see a breakpoint that doesn't have
an associated column, we show the arrow with a dot on the line, rather than
assuming that there'll be an inline indicator arrow.

Use isNaN instead of === NaN

Fixes #83654

Fixes #83702

This reverts commit 53d9528b.