ServiceWorkerURLRequestJob::CreateResponseHeader() is inefficient when there are a large number of response headers. In my local debugging, google servers can send hundreds of experiment ids, so we can end up with ~500 pairs of response headers. HttpResponseHeaders::AddHeader() has a bad performance O(n^2).

We should move away from using HttpResponseHeaders::AddHeader in ServiceWorkerURLRequestJob. mmenke@ suggested that the simplest approach is to just create a single full bogus HTTP/1.x response headers string, and pass that into HttpUtil::AssembleRawHeaders, and then pass the result of that to HttpResponseHeaders().  Or, perhaps better, do what the disk cache does and create a pickle from the HttpResponseHeaders, and use that to re-constitute the headers (The latter won't work with old ServiceWorker cache files, presumably, which wouldn't be great)


An example of such a header is:
["x-google-experiment"] = "17259,750339,750721,751571,1351459,1351563,1351587,1351702,1351810,1351830,1351860,1351903,1352026,1352048,1352081,1352114,1352174,1352193,1352206,1352219,1352241,1352247,1352274,1352277,1352293,13523"...


----------------------------------------------------------
void ServiceWorkerURLRequestJob::CreateResponseHeader(
    int status_code,
    const std::string& status_text,
    const ServiceWorkerHeaderMap& headers) {
  // TODO(kinuko): If the response has an identifier to on-disk cache entry,
  // pull response header from the disk.
  std::string status_line(
      base::StringPrintf("HTTP/1.1 %d %s", status_code, status_text.c_str()));
  status_line.push_back('\0');
  http_response_headers_ = new net::HttpResponseHeaders(status_line);
  for (ServiceWorkerHeaderMap::const_iterator it = headers.begin();
       it != headers.end();
       ++it) {
    std::string header;
    header.reserve(it->first.size() + 2 + it->second.size());
    header.append(it->first);
    header.append(": ");
    header.append(it->second);
    http_response_headers_->AddHeader(header);  <-- Can we avoid this?
  }
}
----------------------------------------------------------