* Correctly report nullability of tuples in the semantic model:
- Store the original tuple literal rather than just its type
- Visit the original literal without diagnostics only in the nullability walker to understand its type
- Update the generator to allow force visiting fields in the nullable re-writer
- Update type checker to allow checking of converted types too
- Update various tuple tests to be correct
- Add extra tests

Support nullable variance in interface and partial method implementation.

Closes #35227.

* Add optional [warnings | annotations] keyword to nullable directive

* pragma warning <setting> nullable -> nullable <setting> warnings

* Implement /nowarn:nullable

Move to a different lowering from for nullable value types to work around a bug in TransformCompoundAssignmentLHS. Addressed PR feedback.

Fixes #36018.

Replace `dynamic` wih `object` when substituting constraints.

Fixes #36276.

Remove support for explicit `object` generic type constraints

* Dont try and check tuple conversions in error cases:
- When checking an identity converison, also check for a bound tuple if the type isn't a tuple as we might have a bound tuple with an error type
- When checking a tuple conversion, don't GenerateConversion in the error case as we don't have a target type to generate against
- Update tests

* Update implicity declared type expression nullability:
- When updating the implicit type in a declaration also update the bound type declaration nullability
- Record nullability for inferred foreach variable
- Add nullable public api tests
- Add nullable tests

* Learn from comparisons to non-constant, non-null values

* Update comment

* Remove unused local function

* Remove another unused local function

* Check for null constant before trying to learn from notNull == maybeNull. Add tests

* More tests, more cleanup

* Add conditional access with || test

* More test updates from feedback

* Cleanup

* Change some locals in tests to parameters

Add initial support for 'notnull' generic type constraint.

See https://github.com/dotnet/csharplang/blob/master/meetings/2019/LDM-2019-05-15.md
Also fixes #36005.

* Infer nested nullability for implicitly typed deconstruction variable:
- When the left hand side is 'var', re-infer the type from the visited right hand side
- Update and add tests

* Allow nullable variance for in/out params when converting method group to delegate

* Fix diagnostic messages for mismatched out parameter nullable variance

* Improve delegate in parameter variance tests

* Add IdentityConversion_DelegateInParameter test

Support re-abstraction of interface members in derived interfaces

Refactored parameter names, updated tests, and added assertions for
assumptions.

Refactored some common code into a helper and corrected some small nits.

Refactors out ApplyConversion into a more aptly named VisitConversion. Adjusts the parameters to take the node being converted, and optionally the original boundconversion. When the boundconversion is passed it, the calculated nullability is now tracked through nested levels of conversions between the top-level boundconversion and the underlying converted node. Removed TrackInferredTypesThroughConversions, as this is now being handled by VisitConversion.

This PR additionally fixes a conversion issue where user-defined conversions that added nullability were not then marking the result type of that conversion as nullable.

* Nullable Deconstruction in foreach loops:
- Visit deconstruction in foreach loops, rather than handling it directly
- Add tests