Implement Display for sketches

[tisonkun]

Good point. I wonder if we'd have a tracking issue to impl fmt::Display for all sketches when applicible. This is what datasketches-java does IMO, but the format can be considered once more.

cc @notfilippo @ariesdevil @ZENOTME

Originally posted by @tisonkun in #204 (comment)

[zhangxinyao88]

I'd like to work on this. To keep the first PR cohesive, I propose starting with the cardinality sketches: HllSketch, HllUnion, CpcSketch, and CpcUnion.

I'll implement summary-only Display output, using the Java summaries as a reference, and add tests for both empty and populated sketches. If that scope sounds good, I'll take it.

[tisonkun]

Go ahead. I'm glad to see a demo to help decide.

[tisonkun]

Thank @zhangxinyao88 for #232.

This hints to me about one more issue:

In Rust, there is a loose contract:

1. If a type implements both FromStr and Display, the following code snippet should pass:

assert_eq!(o, FromStr::from_str(o.to_string()))

2. The result of Display should be stable within one compatible version range.

However, the current printing format looks like:

HLL Sketch Summary:
  lg config k       : 12
  target type       : Hll8
  current mode      : List
  lower bound       : 0
  estimate          : 0
  upper bound       : 0

Seems more like a changable diagnostic report.

cc @leerho what is the design principle of these formats in datasketches-java. Are they mainly for debugging or diagnostics?