feat(net): correct per-response size metric to avoid capacity/empty-block inflation

[phrwlk]

This change fixes downloaders.bodies.response.response_size_bytes in BodiesRequestFuture::try_buffer_blocks to reflect only the sizes of actual bodies received:

• Initialize total_size to 0 instead of using bodies.capacity() * size_of::<C::Body>().
• Do not add any size for empty headers.
• Sum only next_body.size() for non-empty bodies.

Rationale:

• The metric is documented as the size (in bytes) of an individual bodies response. Including the input Vec capacity reflects local allocation details and double-counts memory since BlockBody::size() already accounts for internal capacities. Adding a pseudo-size for empty headers further inflates the metric while response_length counts only real bodies, skewing dashboards (e.g., average size = size/length).
• Elsewhere we include capacities only for buffer footprint metrics (e.g., buffer_bodies_response), not for per-response sizing. This change aligns semantics with documentation and Grafana usage, yielding accurate and comparable observability without affecting functional behavior.