Migrate to Postcard

[sffc]

We currently use Bincode for machine-readable data. Postcard is an alternative Serde plugin that trumpets the following goals:

1. Design primarily for #![no_std] usage, in embedded or other constrained contexts
2. Support a maximal set of serde features, so postcard can be used as a drop in replacement
3. Avoid special differences in code between communication code written for a microcontroller or a desktop/server PC
4. Be resource efficient - memory usage, code size, developer time, and CPU time; in that order
5. Allow library users to customize the serialization and deserialization behavior to fit their bespoke needs

I have found previously that Postcard has a much smaller code size footprint than Bincode (#78 (comment), #866 (comment)), and others have found that it has big wins in memory usage (see here). The second analysis finds that the zlib data size of Postcard is a bit smaller and the CPU performance is a bit slower than Bincode (3.7533 ms for Postcard versus 3.4132 ms for Bincode).

Migrating to Postcard will also help us with the no_std efforts (see #812).

So I would like to propose that we migrate to Postcard for 0.4. If desired, we could keep Bincode around as a feature living alongside Postcard and JSON. I have a slight preference for removing Bincode, simply for the maintenance cost, but I can keep it if people want it.

[zbraniecki]

I'm in support of that move. I second the slight preference to remove Bincode, since it doesn't serve any known additional purpose over Postcard.

[sffc]

2021-07-22:

• @iainireland - It makes sense to be opinionated about the best option for ICU4X
• @gregtatum - +1 to @iainireland

Consensus: Adopt Postcard as proposed above.