fix(nextflow): don't mark workspace scan flushed when both probes fail

_flush_deferred_workspace_scan sets _workspace_scan_flushed unconditionally after its
two completion() flush attempts, even when both raised. A later find_referencing_symbols
call then silently skips the flush and answers from a possibly-incomplete workspace AST
cache for the rest of the session, with no exception and no retry.

Fixes #1871

test/solidlsp/test_nextflow_workspace_scan_flush.py

@@ -0,0 +1,65 @@
+"""Regression test for the Nextflow deferred-workspace-scan flush flag.
+
+``_flush_deferred_workspace_scan`` sends two ``completion`` requests purely to force the
+server's debounced workspace scan, discarding their results; ``_workspace_scan_flushed``
+exists only to skip that work once it is known to have happened. If both requests raise,
+nothing was actually flushed and the flag must stay clear so the next call retries,
+without spawning a real language server process.
+"""
+
+from solidlsp.language_servers.nextflow_language_server import NextflowLanguageServer
+
+
+class _FakeCompletionSender:
+    def __init__(self, *, always_raise: bool = False) -> None:
+        self.calls = 0
+        self._always_raise = always_raise
+
+    def completion(self, params: dict) -> None:
+        self.calls += 1
+        if self._always_raise:
+            raise TimeoutError("no response from the Nextflow language server")
+
+
+class _FakeServer:
+    def __init__(self, sender: _FakeCompletionSender) -> None:
+        self.send = sender
+
+
+def _bare_nextflow_ls(sender: _FakeCompletionSender, repo_path: str) -> NextflowLanguageServer:
+    """Create an instance without running __init__ (no JVM, no process), setting only the
+    state the flush machinery touches; same technique as test_typescript_timeout_policy.py.
+    """
+    ls = object.__new__(NextflowLanguageServer)
+    ls.repository_root_path = repo_path
+    ls._workspace_scan_flushed = False
+    ls.server = _FakeServer(sender)
+    return ls
+
+
+def test_flag_stays_clear_when_both_completion_calls_fail(tmp_path) -> None:
+    sender = _FakeCompletionSender(always_raise=True)
+    ls = _bare_nextflow_ls(sender, str(tmp_path))
+
+    ls._flush_deferred_workspace_scan("main.nf")
+
+    assert sender.calls == 2
+    assert ls._workspace_scan_flushed is False
+
+    # a later call must retry, not silently skip forever
+    ls._flush_deferred_workspace_scan("main.nf")
+    assert sender.calls == 4
+
+
+def test_flag_is_set_when_a_completion_call_succeeds(tmp_path) -> None:
+    sender = _FakeCompletionSender(always_raise=False)
+    ls = _bare_nextflow_ls(sender, str(tmp_path))
+
+    ls._flush_deferred_workspace_scan("main.nf")
+
+    assert sender.calls == 2
+    assert ls._workspace_scan_flushed is True
+
+    # once flushed, a later call is a no-op
+    ls._flush_deferred_workspace_scan("main.nf")
+    assert sender.calls == 2