fix(web-client): js_u64_to_u64 throws instead of panicking on out-of-range BigInt

[ygd58]

js_u64_to_u64 (the Node.js-side JsU64 -> u64 conversion, used at every JS-facing bigint parameter in this crate) panicked whenever the caller passed a BigInt that was negative or >= 2^64 - napi's BigInt has no such constraint at the type level, so this is ordinary, easily-reachable JS input, not a contrived edge case.

Scope

This was a single, widely-shared helper - 33 call sites across 11 files - so a negative or oversized amount/index/timestamp anywhere in this SDK's surface (Felt, Word, FungibleAsset, TransactionStatus.committed, TransactionProver.newRemoteProver's timeout, or any mint/send/swap transaction request builder's amount) crashed the whole module instead of throwing a catchable error.

Fix

Changes js_u64_to_u64 to return Result<u64, JsErr>. Every call site already sat inside a function returning Result<_, JsErr>, so most just gained a ?. Two call sites (TransactionProver::new_remote_prover, TransactionStatus::committed) previously returned their value directly (no Result) and needed their own signature changed - verified neither has another Rust-side caller in this crate. MerklePath::verify and SparseMerklePath::verify return plain bool and already collapse the native call's own Result into false on failure; an invalid index now takes the same false path rather than becoming a Result-returning API change.

Test plan

Added crates/web-client/test/js_u64_out_of_range.test.ts (6 cases) covering Felt, Word, FungibleAsset and TransactionStatus with negative and >=2^64 BigInt inputs, plus one confirming valid input is unaffected.

Ran locally, not just CI:

cargo build -p miden-client-web --no-default-features --features nodejs,testing --release
SKIP_WEB_SERVER=1 pnpm exec playwright test --project=nodejs --workers=1

Build: clean. Full suite: 176 passed, 25 skipped, 0 failed (201 total) - all 6 new cases included, no regressions.