Because log `Record`s are now being passed by ownership to `log`,
the bindings get quite annoyed that there's a lifetime hanging
around. We already don't use this lifetime in bindings (the
`FormatArgs` is converted to a string and stored on the heap), so
we can just drop the lifetime, even though it requires some
macro'ing of the struct definition to do so.

There's not a lot of reason for downstream users to use the
`WithContext` wrapper, it mostly exists for our own downstream
crates anyway, and dealing with lifetimes in bindings isn't super
practical, so simply no-export it.

The bindings cannot express lifetimes, so exposing any field which
is a reference (and not `clone`-able, at least for garbage
collected language bindings) is unsafe for those expecting a
high-level interface.

Thus, we simply no-export the `RouteHopCandidate` inner struct
fields which are references (there are relevant accessors for them
anyway).

In `three_hop_blinded_path_success`, the nodes in the test ended up
at radically different block heights after channel opening. At that
point, if the CLTV randomization is done slightly different the
test payment may fail, which we fix here by ensuring all nodes are
at the same height before we go to send a payment.

...as an arg to `Router`. Passing an `EntropySource` around all
the time is a bit strange as the `Router` may or may not actually
use it, and the `DefaultRouter` can just as easily store it.

The top-level module is only a few hundred lines, so there's not a
lot of reason to hide the `GraphSyncError` in its own module.
Instead, we simply move it to the top-level `lib.rs`, which doesn't
change the public API as it was previously re-exported at the top
level.

As it does the same thing as a derived `Debug` does anyway.

In 26c1639 (and later in
50c55dc) we switched to using
`Default::default()` to initialize `()` for scoring parameters in
tests. A number of `()`s slipped back in recently, which we replace
here.