Add general error type

[ZENOTME]

For now, there is only SerdeError:

datasketches-rust/datasketches/src/error.rs

Line 24 in 8855aa6

pub enum SerdeError {

In some cases, we hope to return an error, but not in the serde case. So a general error type is need

[tisonkun]

See #27 (comment)

I think we may leverage some practices in OpenDAL's Error type. cc @Xuanwo

[tisonkun]

In some cases

@ZENOTME Also, it would be helpful if you could call out those specific cases now so that we can build a pattern to resolve concrete issues rather than a hypothetical one.

[tisonkun]

I can consider three ErrorKind now:

1. Serde error. Mainly for deserialization when hitting invalid data.
2. Argument error. We have some argument contracts on the construction of sketches. Now we panic if the contract doesn't meet. But we may provide a fallible constructor try_new and return Error { Kind: ConfigInvalid } on unexpected arguments.
3. State error. More common internal error. For example, if the internal state is invalid at some point. datasketches-java has SketchesStateException for several cases.

@leerho @freakyzoidberg @AlexanderSaydakov I'd like to learn from other impls' experience with error handling and see how to integrate that experience in a Rust-idiomatic way.

[tisonkun]

And the general Error type would look like:

pub struct Error {
    kind: ErrorKind,
    message: String,
    context: Vec<(&'static str, String)>,
    backtrace: Option<Box<Backtrace>>,
}

See also https://github.com/apache/opendal/blob/18fa9fe953f8579c9c4685a00d2524a68dad7bbb/core/core/src/types/error.rs#L223C1-L233C2

Other fields may be added when we find them useful in datasketches' scope.

[ZENOTME]

In some cases

@ZENOTME Also, it would be helpful if you could call out those specific cases now so that we can build a pattern to resolve concrete issues rather than a hypothetical one.

E.g. In theta sketch, the builder is better to return an error rather than asserting.

impl ThetaSketchBuilder {
    /// Set lg_k (log2 of nominal size k)
    pub fn set_lg_k(mut self, lg_k: u8) -> Self {
        // # TODO
        // Return as error
        assert!(
            (MIN_LG_K..=MAX_LG_K).contains(&lg_k),
            "lg_k must be in [{}, {}], got {}",
            MIN_LG_K,
            MAX_LG_K,
            lg_k
        );
        self.lg_k = lg_k;
        self
    }

I just checked: #27 (comment) , and it's reasonable to keep it as an assert, and the contract is: panic for a lg_k not in [MIN_LG_K, MAX_LG_K] and add the related comment. After we introduce the Error type, we can have a try_xxx version to return the error when the contract is violated.