Optimise block fetch decision logic in bulk sync mode

The block fetch decision logic shows up high in profile output.

@dcoutts:
> The overall point is that we're making fetch decisions too frequently. Each decision is probably appropriately expensive, but we're making them too frequently, at least in bulk fetch mode.
>
> I suspect we're requesting rather short ranges of blocks, and doing so too frequently. It'd be better to make fewer decisions, and fetch larger ranges of blocks.