Avoid double JSON serialization in batchAddRequests #972

Follow-up from the Python client's equivalent fix — see apify/apify-client-python#953 and the discussion there. The JS client has the same inefficiency.

In batchAddRequests, every request is JSON-serialized twice:

1. Once to measure payload size when splitting into batches — sliceArrayByByteLength (src/utils.ts) calls JSON.stringify on the whole slice, and on the splitting path on each request individually.
2. Once more when sending — the batch is passed as data to httpClient.call, and the serializeRequest axios interceptor (src/interceptors.ts) stringifies the same objects again to build the request body.

That's redundant CPU work in a hot path — Crawlee's RequestQueue goes through this method.

Suggested fix, mirroring apify/apify-client-python#953: serialize each request once up front, split batches by the serialized byte lengths (counting the , separators and the [] brackets), and send the pre-assembled JSON string as the body with an explicit content-type: application/json header so the interceptor doesn't re-serialize it. As a side effect this also fixes the current measurement slightly undercounting the actual body: it accounts for the [] brackets but not the commas between items (harmless today thanks to the safety buffer, but fixed for free by this approach).

✍️ Drafted by Claude Code