Split integration tests to different targets

Running all integration tests as part of one single 'replication_it'
target does not cope well with the addition of extra tests, because it
is bound to take longer and longer, eventually hitting any test timeout
threshold.

Splitting integration tests into different targets avoids timeout
failures and also provides additional benefits, such as:

- Better understanding of test failures
- More efficient utilization of bazel build outputs and remote caching,
  effectively making tests execution faster.

Bug: Issue 13909
Change-Id: Ifc6cce9996d3a8a23ec2a66c377978205fb6680f

BUILD

diff --git a/BUILD b/BUILD
index 72a3fc8..1bf917d 100644
--- a/BUILD
+++ b/BUILD
@@ -32,18 +32,16 @@
     ],
 )
 
-junit_tests(
-    name = "replication_it",
-    srcs = glob([
-        "src/test/java/**/*IT.java",
-    ]),
+[junit_tests(
+    name = f[:f.index(".")].replace("/", "_"),
+    srcs = [f],
     tags = ["replication"],
     visibility = ["//visibility:public"],
     deps = PLUGIN_TEST_DEPS + PLUGIN_DEPS + [
         ":replication__plugin",
         ":replication_util",
     ],
-)
+) for f in glob(["src/test/java/**/*IT.java"])]
 
 java_library(
     name = "replication_util",