refactor(api): replace response copy wrappers with Jetty content count

[halibobo1205]

What does this PR do?

Replaces response-body copying for HTTP traffic metrics with Jetty's native Response.getContentCount(). Removes CharResponseWrapper, ServletOutputStreamCopy, and the unused response wrapping in HttpApiAccessFilter.

Why are these changes required?

The old wrappers retained response bodies in memory solely to measure their size. Full-node wallet requests could pass through two copies, and bulk writes fell back to a per-byte loop. Jetty already maintains a byte count, including content still in its output buffer, so these copies are unnecessary. HttpApiAccessFilter built a wrapper it never read, on four HTTP services — three of which register no interceptor at all.

Using Jetty's native response writer also fixes incomplete responses caused by bytes remaining in the old wrapper's unflushed OutputStreamWriter buffer. Unflushed print() calls could lose an entire small response or the trailing bytes of a larger response. This is latent rather than live: every servlet under services/http ends with println, and the one non-println writer call runs upstream of any wrapper.

Before/after checks found matching response bodies and byte counts for the tested synchronous production write patterns, including println, JSON-RPC output-stream writes, and UTF-8 responses.

Metrics are still recorded after the filter chain returns. They measure response-body bytes written to Jetty, excluding HTTP headers and chunk framing; they do not guarantee delivery to the client. Existing exclusions for disabled-API and lite-node responses remain. Requests without a Jetty base request record zero bytes, and the new lookup introduces no cast. Production filter registration and request parsing are unchanged.

This PR has been tested by:

• Unit Tests
  • Real-Jetty tests cover println, unflushed print, output-stream writes with Content-Length,
    UTF-8, a body past the output buffer, and HTTP 400, using the access-filter-before-interceptor
    ordering of full-node wallet endpoints. An outer completion latch keeps the metric assertions
    deterministic. Unit tests cover the non-Jetty fallback and propagation to both traffic meters.
  • On the pre-change classes, the print and large-body tests fail; mutation testing confirms the
    remaining assertions are not vacuous.
  • Focused tests passed.
• Manual Testing
  • Test by private chain.

Follow up

Heads-up for whoever lands #6923: it removes the Codahale MetricsUtil.meterMark calls from this
same method, so the two will conflict textually. The byte count itself is still needed afterwards
for the Prometheus tron:http_bytes histogram.