There is no conversion between ref like types and either object or generic parameters (even if constrained to struct) in either direction. This change enforces that rule and adds tests around pattern matching, which exercises this rule.

Most constructs already had tests, but I added a few more just in case.
Although Span return in iterators is currently impossible because no
ref-struct capable type can be an iterator, I added a safety check in
case we allow it later.

Fixes #22305

Temporarily moves checks for `partial` back into the parser.

Fixes #22379

Fixes #18348

* updated local decl

* put IVariableDeclarationStatement back to for loop before.

* put variable declaration syntax in variable declaration statement for for loop

* updated tests

* changed caller to call different APIs to create variable declaration when it can.

* changed to return IVaraibleDeclaration from CreateVariableDeclaration method

* addressed PR feedbacks

* Clean up

* Test fixes

* Fix failing test

- "in" argument matches only "in" parameters (as opposed to unmodified case that can also match byval parameters)
- "in" requires that the argument is an lvalue
- "in" does not allow conversions at the call-site other than identity conversions
- "in" sets an exact bound in generic method type inference
- "in" cannot be stack-spilled by value

Fixes:#22306

Fixes #22006

* Create tests to cover test plans #19216 and #20127

* clean up

* Improve consistency of handling constant null switch expression
Fixes #20210
* Rationalize reachability for pattern switch.
If a single case could handle all inputs, e.g. has the
same constant value, a wildcard pattern, etc, then
the default label is considered unreachable.
The concept of reachability, which takes the value of the
switch expression into account and can produce warnings,
is separate from the concept of subsumption which takes
the full set of patterns into account and can produce errors.
* Document breaking change in pattern switch.
Alse change formatting as suggested in code review.