chore(java): refactor error handling

[LuQQiu]

Refactor the error handling to map errors (with an ErrorExt trait) following Lance Python convention

[LuQQiu]

@eddyxu @QianZhu PTAL, please mainly take a look this class.
The others are mainly moving classes around, and refactor it to use the new error handling approach

[QianZhu]

@LuQQiu can we add tests to capture those errors? Specifically, whether the clear error message is presented.

[eddyxu]

We usually just name this as

type Result<T> = std::result::Result<T, Error>`


Struct Error {
...}

So, within this crate, it still reads in a style that is consistent with a Rust project.

[LuQQiu]

This is to mimic the Python side, turn Result finally to a special end Result

[broccoliSpicy]

as @wjones127 pointed out in #2316, to get more precise error info, it might be better to put a BoxedError here. like

#[derive(Debug)]
pub struct JavaError {
    source: BoxedError,
    java_class: JavaExceptionClass,
}

BoxedError is defined here:
BoxedError

I haven't read much of the Lance codebase, correct me if I'm wrong

[LuQQiu]

@broccoliSpicy Good to know about the BoxedError and its benefits! Thanks for the knowledge transfer!
For java side, this JavaError is a last mile error to the user, which translates the LanceError into Java Error with string message.
The information in the BoxedError will be included in this String message
e.g. In lance Error,
#[snafu(display("Dataset at path {path} was not found: {source}, {location}"))]
DatasetNotFound {
path: String,
source: BoxedError,
location: Location,
},
on java side we directly call LanceError.toString() to get the message with source information.

[broccoliSpicy]

@LuQQiu glad to know how error message is transfered from lance to jave, thanks!

[LuQQiu]

@QianZhu PTAL, thanks!

[LuQQiu]

@QianZhu Added some test cases
while adding the test cases, found out some unsafe operations so

1. Add the ReentrantReadWriteLock for dataset and scanner objects
2. Avoid clone the Rust JNI object, following the safety requirement at https://docs.rs/jni/latest/jni/struct.JNIEnv.html#method.set_rust_field

[eddyxu]

This is first time i saw the conversion on Result instead of Error. Why do we do this?

[LuQQiu]

Mimic the Python side to reduce the error conversion headache, we do not need to list out all the errors and create the message for each error again, makes the error conversion much easier (keep the original info while can convert to target java exception)

[eddyxu]

Lets just name this to Error?

[LuQQiu]

This is to mimic PyResult and PyErr lol
represent a special kind of end result/error