[chunks] Move chain_head maintenance into ShardsManager

[robin-near]

• chain_head, which used to be passed into the ShardsManager with the process_partial_encoded_* calls, is now kept by the ShardsManager, which is required to send chunk management network messages directly to the ShardsManager. Whenever the chain_head is updated, the Client will provide the new chain_head to the ShardsManager. This makes chain_head updating asynchronous, but that is OK, because this is only used to opportunistically validate orphan chunk headers, and the accuracy of the chain_head does not affect correctness.
• Note that "header_head", which is different from chain_head (i.e. it only waits for a valid block header but not the entire valid block), is still passed in; this is used for the chunk requesting path. This is not a problem, because the request path is initiated by the Client (upon receiving a block with missing chunks), except that we also pass in a header_head when resending requests; the use of header_head for this context does not require an accurate header_head, and this will be handled in a later PR to read from the store directly.
• This PR also moves response handling, forward handling, and incomplete chunks checking to the ShardsManager, bypassing the Client completely. It also makes some functions no longer pub.

[mzhangmzz]

Linking it with #7662

[matklad]

Add comment that this is a "best effort" and updated asynchronously, and point to where the source of truth is?

I think it's useful to explicitly mark "this is not source of truth" bit

[robin-near]

Agreed! Added a comment.