Client: hide pointer indirection.

[zenhack]

This changes the APIs so that everything operates on Client instead of
*Client. This will smooth over a couple things I've run into, mainly
around being able to implement interfaces where a method needs to
re-assign the receiver.

I'm open to debate as to whether this is a good change. Copying my message from matrix for posterity:

@lthibault: How do you feel about changing the API to use Client instead of *Client everywhere (and probably hide the pointer indirection inside the struct)? I think this would make a lot of stuff esp. with generics cleaner.

E.g. on my generics branch right now I have:

package capnp

// The TypeParam interface must be satisified by a type to be used as a canpproto
// type parameter. This is satisified by all capnproto pointer types. T should
// be instantiated by the type itself; in type parameter lists you will typically
// see parameter/constraints like T TypeParam[T].
type TypeParam[T any] interface {
	// Convert the receiver to a Ptr, storing it in seg if it is not
	// already associated with some message (only true for Clients and
	// wrappers around them.
	EncodeAsPtr(seg *Segment) Ptr

  // Decode the pointer as the type of the receiver. Generally,
  // the reciever will be the zero value for the type.
  DecodeFromPtr(p Ptr) T
}

...but I'd like to drop the T parameter so it would just be:

type TypeParam interface {
   EncodeAsPtr(*Segment) Ptr
   DecodeFromPtr(Ptr)
}

...but I can't do the latter because I would have to implement DecodeFromPtr on **Client, since it needs to modify the pointer. But you can't define methods on double-indirect pointers like that.
Returning the T is kindof an awful hack so we can just define it on *Client instead.
But if we hide the pointer indirection inside the struct, then we can define the method on *Client, and it can modify it in place.

Despite the size of the diff, I don't think this will be horribly disruptive to downstream code; for ocap-md I just needed to re-run the schema compiler and it just worked. Changes should be pretty mechanical in the cases where you are working with raw Clients.

[zenhack]

Looks like the tests are passing again here too.

[lthibault]

LGTM