When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

Also change the lowering of splatting to use this new intrinsic directly,
thus fixing #26001.

When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

Also change the lowering of splatting to use this new intrinsic directly,
thus fixing #26001.

When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

Also change the lowering of splatting to use this new intrinsic directly,
thus fixing #26001.

When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

Also change the lowering of splatting to use this new intrinsic directly,
thus fixing #26001.

)

When using the Casette mechanism to intercept calls to _apply,
a common strategy is to rewrite the function argument to properly
consider the context and then falling back to regular _apply.
However, as showin in JuliaLabs/Cassette.jl#146,
this strategy is insufficient as the _apply itself may recurse into
various `iterate` calls which are not properly tracked. This is an
attempt to resolve this problem with a minimal performance penalty.
Attempting to duplicate the _apply logic in julia, would lead to
code that is very hard for inference (and nested Cassette passes to
understand). In contrast, this simply adds a version of _apply that
takes `iterate` as an explicit argument. Cassette and similar tools
can override this argument and provide a function that properly
allows the context to recurse through the iteration, while still
allowing inference to take advantage of the special handling of _apply
for simple cases.

Also change the lowering of splatting to use this new intrinsic directly,
thus fixing #26001.