Cuckatoo size shift upgrade schedule

[ignopeverell]

Fixes #2001

[tromp]

31 + (height / YEAR_HEIGHT)
is an exponential size increase; not the linear one we want.
what we want is 31 + log_2(height / YEAR_HEIGHT)

but instead I propose to leave DEFAULT_MIN_EDGE_BITS the constant it was,
and rely on the graph weight function to codify the phasing out schedule,
since we need that anyway.

[ignopeverell]

@tromp just pushed an update that moves the schedule into graph_weight. It could be written more succinctly or elegantly but I wanted to first have an agreement on that method before finicking.

[tromp]

I find it clearer in term of an explicitly computed expiry height, e.g.

    let bits_over_min = edge_bits - global::min_edge_bits();
    let expiry_height = (1 << bits_over_min) * YEAR_HEIGHT;
    if height > expiry_height {
            xpr_edge_bits =
                    xpr_edge_bits.saturating_sub(1 + (height - expiry_height) / WEEK_HEIGHT);
    }

[ignopeverell]

It is clearer. Finished up the PR and have the test passing now. Should be ready for merge.