fix(compute-provider): ComputeManager accepts batch_size Some(0) and then panics

[Toby1009]

ComputeManager::new accepts batch_size: Some(0) without validation, and start() then panics
inside slice::chunks.

Reproduce

use e3_compute_provider::{ComputeInput, ComputeManager, ComputeProvider, FHEInputs};

struct Dummy;
impl ComputeProvider for Dummy {
    type Output = ();
    fn prove(&self, _: &ComputeInput) -> Self::Output {}
}

fn processor(_: &FHEInputs) -> Vec<u8> { Vec::new() }

fn main() {
    let inputs = FHEInputs { ciphertexts: vec![], params: vec![] };
    let mut m = ComputeManager::new(Dummy, inputs, processor, true, Some(0));
    println!("constructor accepted batch_size = Some(0)");
    let _ = m.start();
}

Output:

constructor accepted batch_size = Some(0)

thread 'main' panicked at crates/compute-provider/src/compute_manager.rs:84:14:
chunk size must be non-zero

Where

ComputeManager::new stores the value as given (src/compute_manager.rs:30-47).
start_parallel resolves it with unwrap_or(2) and passes it straight to chunks (:78, :84),
before any FHE work, so the panic happens on the first parallel run regardless of the input.

Reachability

No caller in this repository hits it. start_parallel has no external call sites, and all three
ComputeManager::new call sites pass use_parallel: false and batch_size: None
(crates/support-scripts/dev/src/main.rs:50, crates/support/host/src/lib.rs:431, :470).

ComputeManager is public API of the published e3-compute-provider crate, and
crates/compute-provider/Readme.md documents both parameters, so a downstream user can reach it.

Note on the documentation

The crate README claimed batch_size "must be a power of 2", which nothing enforced. #1813 corrects
that text to describe the current behaviour. This issue is about the code: whether the constructor
should reject Some(0), and whether the power-of-two constraint the README described was intended
and should be enforced too.

Happy to open a PR once the intended behaviour is settled.

[Toby1009]

Fixed by #1821, which removed the parallel path rather than validating the argument.

On main at 3c7172c8, ComputeManager::new takes three parameters
(crates/compute-provider/src/compute_manager.rs:25-27), and use_parallel, batch_size, the
.chunks() call, and the crate's rayon dependency are all gone. The only remaining mentions in
the crate are two lines of stale documentation (Readme.md:87-88), which #1808 tracks and #1813
corrects.

Closing.