feat: YAML pipeline template autocomplete (AzDO + GHA)

- Add scripts/sync_pipeline_templates.py — scans LRU AzDO and GHA template
  repos; outputs unified pipeline_templates_catalog.json (48 templates: 45
  AzDO + 3 GHA)
- Add scripts/template_sources.yml — source config (AzDO alias, GHA org)
- Add pipeline_templates_catalog.json — baked catalog (49 KB)
- Add ilsp/yaml_lsp/catalog.py — PipelineTemplateCatalog with completion item
  generators for template paths, param names, allowed values, GHA inputs
- Add ilsp/yaml_lsp/proxy.py — async WS↔TCP bridge with LSP frame buffering,
  per-connection document tracking, AzDO/GHA context detection, and completion
  injection (LRU items sortText 0_, standard items downgraded to 9_)
- Wire yaml_ws_handler into server.py (replaces raw _ws_proxy call)
- Load PipelineTemplateCatalog at startup; reload + health report template count
- Update push_catalogs.sh to push pipeline_templates_catalog.json
- Update Dockerfile to bake pipeline_templates_catalog.json as image fallback
- Add tests/test_yaml_catalog.py (14 tests) + tests/test_yaml_proxy.py (18 tests)
  All 67 tests green

ilsp/server.py

@@ -24,6 +24,8 @@ from aiohttp import web
 from .python_lsp.catalog import PypiCatalog
 from .bicep_lsp.modules import BicepModuleCatalog
 from .bicep_lsp.proxy import serve_bicep
+from .yaml_lsp.catalog import PipelineTemplateCatalog
+from .yaml_lsp.proxy import yaml_ws_handler
 
 logger = logging.getLogger(__name__)
 
@@ -131,17 +133,26 @@ async def _build_app() -> web.Application:
             "bicep_modules": len(BicepModuleCatalog._modules),
             "iac_source_modules": len(BicepModuleCatalog._iac),
             "yaml_lsp": bool(shutil.which("yaml-language-server")),
+            "pipeline_templates": PipelineTemplateCatalog.template_count(),
         })
 
     async def reload(_: web.Request) -> web.Response:
-        before = len(BicepModuleCatalog._modules)
+        before_bicep = len(BicepModuleCatalog._modules)
+        before_tmpl = PipelineTemplateCatalog.template_count()
         BicepModuleCatalog.load()
-        after = len(BicepModuleCatalog._modules)
-        logger.info("Catalog reloaded: %d → %d modules", before, after)
+        PipelineTemplateCatalog.load()
+        after_bicep = len(BicepModuleCatalog._modules)
+        after_tmpl = PipelineTemplateCatalog.template_count()
+        logger.info(
+            "Catalog reloaded: bicep %d→%d  templates %d→%d",
+            before_bicep, after_bicep, before_tmpl, after_tmpl,
+        )
         return web.json_response({
             "status": "reloaded",
-            "bicep_modules_before": before,
-            "bicep_modules_after": after,
+            "bicep_modules_before": before_bicep,
+            "bicep_modules_after": after_bicep,
+            "pipeline_templates_before": before_tmpl,
+            "pipeline_templates_after": after_tmpl,
         })
 
     async def python_ws(request: web.Request) -> web.WebSocketResponse:
@@ -151,7 +162,7 @@ async def _build_app() -> web.Application:
         return await _ws_proxy(request, "127.0.0.1", BICEP_LSP_PORT)
 
     async def yaml_ws(request: web.Request) -> web.WebSocketResponse:
-        return await _ws_proxy(request, "127.0.0.1", YAML_LSP_PORT)
+        return await yaml_ws_handler(request, YAML_LSP_PORT)
 
     app.router.add_get("/health", health)
     app.router.add_post("/reload", reload)
@@ -169,6 +180,7 @@ async def main_async() -> None:
 
     logger.info("Pre-warming catalogs…")
     BicepModuleCatalog.load()
+    PipelineTemplateCatalog.load()
     await PypiCatalog.start_background_refresh()
 
     # LSP servers run internally on localhost — not exposed outside the container