Update TranslationModel to boost speed

Towards #77.

This updates TranslationModel to use the new
Service::translateMultiple(...) API to efficiently pack batches by
compiling multiple Requests and using only using one translate-dispatch
call for all Requests.

Due to changes in batching/ordering and consequently difference in
approximations after quantization, the outputs are now different from those
earlier (breaking bergamot-translator-app regression-tests) but correct
nonetheless (regression-tests output now need updating).

src/translator/TranslationModel.cpp

@@ -9,6 +9,7 @@
 // All local project includes
 #include "TranslationModel.h"
 #include "translator/parser.h"
+#include "translator/response.h"
 #include "translator/service.h"
 
 TranslationModel::TranslationModel(const std::string &config,
@@ -21,31 +22,25 @@ TranslationModel::~TranslationModel() {}
 std::vector<TranslationResult>
 TranslationModel::translate(std::vector<std::string> &&texts,
                             TranslationRequest request) {
-  // Implementing a non-async version first. Unpleasant, but should work.
-  std::promise<std::vector<TranslationResult>> promise;
-  auto future = promise.get_future();
 
   // This code, move into async?
   std::vector<TranslationResult> translationResults;
-  for (auto &text : texts) {
-    // Collect future as marian::bergamot::TranslationResult
-    auto intermediate = service_.translate(std::move(text));
-    intermediate.wait();
-    auto marianResponse(std::move(intermediate.get()));
-
+  std::vector<marian::bergamot::Response> responses =
+      service_.translateMultiple(std::move(texts), request);
+  for (auto &response : responses) {
     TranslationResult::SentenceMappings sentenceMappings;
-    for (size_t idx = 0; idx < marianResponse.size(); idx++) {
-      marian::string_view src = marianResponse.source.sentence(idx);
-      marian::string_view tgt = marianResponse.target.sentence(idx);
+    for (size_t idx = 0; idx < response.size(); idx++) {
+      marian::string_view src = response.source.sentence(idx);
+      marian::string_view tgt = response.target.sentence(idx);
       sentenceMappings.emplace_back(std::string_view(src.data(), src.size()),
                                     std::string_view(tgt.data(), tgt.size()));
     }
 
     // In place construction.
     translationResults.emplace_back(
-        std::move(marianResponse.source.text), // &&marianResponse.source_
-        std::move(marianResponse.target.text), // &&marianResponse.translation_
-        std::move(sentenceMappings)            // &&sentenceMappings
+        std::move(response.source.text), // &&response.source_
+        std::move(response.target.text), // &&response.translation_
+        std::move(sentenceMappings)      // &&sentenceMappings
     );
   }