fix(xdr): bound decode limits to prevent memory-exhaustion DoS

[Ayinkx]

Problem

The Fuzz workflow is failing on main. The decode_scval target found an OOM (memory-exhaustion denial of service) in devkit-xdr:

cargo fuzz run decode_scval fuzz/artifacts/decode_scval/oom-acb2aaa1... Error: Fuzz target exited with exit status: 71

Root cause

crates/devkit-xdr/src/lib.rs decoded untrusted XDR with Limits::none(), which sets depth and len to their maximum values — i.e. no limits. In stellar-xdr, variable-length types call consume_len(len) and then allocate �ec![0u8; len]. With no limit, a crafted length prefix requests an enormous allocation.

Fix

Apply bounded reader limits:

• depth: 128 (stack safety)
• len: 16 MiB (bounds total bytes read; a declared length that would exceed this is rejected before allocation)

Tests

ejects_oversized_scval_from_fuzzer_without_exhausting_memory: replays the exact fuzzer input; asserts a structured decode error and no unbounded allocation.

• limits_are_bounded: guards against a regression back to unbounded limits.

Full gate passes: cargo fmt --check, cargo clippy --workspace --all-targets --all-features -- -D warnings, cargo test --workspace.

Fixes the failing Fuzz workflow on main.