Named type converters

[hannahhoward]

Goals

Currently, the bindnode API enables custom conversions, keyed on go types.

This PR enables a new way to key custom conversions: the type name in a schema.

Why is this useful?

• First it enables using standard go types for custom conversions, rather than needing to create a named go type every time you want a converter
• More interestingly, it enables using interface types within structs, lists, and maps -- since the conversion is keyed on the schema type name, the go type isn't needed to setup a custom conversion, and then the convert can create types that satisfy the interface -- this is super useful for more flexible go structs.

Implementation

• For compatibility, this keeps all typed custom converter APIs working, with schema named converters taking preference. This does mean that the config type must become a struct, and therefore a *config where it is passed
• The schema type name is passed to every call for converterFor & converterForType. Not sure what performance implications this may have (mostly in lookups)
• Wrote a test demonstrating equivalent functioning working with named converters, and a test showing how to use it with interfaces (here used to define a closed union type, something I find is useful to represent often)

[rvagg]

cool

[rvagg]

windows ci failures are known flaky garbage, build output "ipldsch_minima.exe" already exists and is not an object file