test: limit doctest parallelism

Due to doctest compilation model they've used a lot of memory leading to
OOMs in CI.

Makefile

@@ -59,11 +59,14 @@ build-release-pkg: build
 
 # XXX: make sure we pass proper flags to cargo test so resulting picodata binary is reused
 # can be reused for python tests without recompilation
-# Note: mock feature is needed for sbroad to compile in test mode
 # Note: tarantool and tlua are skipped intentionally, no need to run their doc tests in picodata
 # Note: gostech-audit-log and picodata-plugin tests simply do not pass:
 #	https://git.picodata.io/picodata/picodata/picodata/-/issues/1084
 #	https://git.picodata.io/picodata/picodata/picodata/-/issues/1085
+# Note: non-doc tests and doc tests are run separately. This is intended to prevent excessive
+# memory usage caused by doc tests compilation model. Doc tests are compiled as part of actual test run.
+# So, each parallel thread lanuched by cargo test spawns full blown compiler for each doctest
+# which at the end leads to OOM.
 .PHONY: test
 test:
 	cargo test $(LOCKED) $(MAKE_JOBSERVER_ARGS) $(CARGO_FLAGS) $(CARGO_FLAGS_EXTRA) $(ERROR_INJECTION) \
@@ -71,7 +74,17 @@ test:
 	  --exclude picodata-plugin \
 	  --exclude sbroad-core \
 	  --exclude tarantool \
-	  --exclude tlua
+	  --exclude tlua \
+	  --tests
+
+	cargo test $(LOCKED) $(MAKE_JOBSERVER_ARGS) $(CARGO_FLAGS) $(CARGO_FLAGS_EXTRA) $(ERROR_INJECTION) \
+	  --exclude gostech-audit-log \
+	  --exclude picodata-plugin \
+	  --exclude sbroad-core \
+	  --exclude tarantool \
+	  --exclude tlua \
+	  --doc -- --test-threads 2
+
 	poetry run pytest $(PYTEST_NUMPROCESSES) $(PYTEST_FLAGS) -vv --color=yes
 
 .PHONY: generate