Optimize doc builds (rapidsai#14856)

cudf docs are generally very slow to build. This problem was exacerbated by the recent addition of libcudf C++ API documentation to the Sphinx build. This PR aims to ameliorate this issue for both local and CI builds by making the following changes:
- The XML parsing logic used to clean up doxygen XML now avoids rewriting files unless they are actually modified. This prevents Sphinx from doing extra work during a second (text) build after the first (HTML) build.
- toctrees on the generated API pages are removed (see https://pydata-sphinx-theme.readthedocs.io/en/stable/user_guide/performance.html#selectively-remove-pages-from-your-sidebar).
- Text docs are disabled in PRs and only occur in nightly/branch builds.

The net result is roughly a halving of the CI run time for the builds (~40 min to ~20 min). Further potential optimizations:
- Reenabling parallel builds. We cannot fully revert rapidsai#14796 until the theme is fixed, but if we can put in a warning filter we could reenable parallelism and have it work on just the reading steps of the build and not the writes. That would still improve performance.
- Better caching of notebooks. [nbsphinx supports caching](https://myst-nb.readthedocs.io/en/latest/computation/execute.html#execute-cache), but there are various caveats w.r.t. 1) local vs CI builds, 2) proper cache invalidation, e.g. when notebook source does not change but underlying libraries do, and 3) forcing rebuilds. Alternatively, we could enable some environment variable that allows devs to turn off notebook execution locally. Making it opt-in would make the default behavior safe while providing an escape hatch for power users who want the builds to be fast.

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - AJ Schmidt (https://github.com/ajschmidt8)
  - Lawrence Mitchell (https://github.com/wence-)

URL: rapidsai#14856

ci/build_docs.sh

@@ -41,19 +41,25 @@ popd
 rapids-logger "Build Python docs"
 pushd docs/cudf
 make dirhtml
-make text
-mkdir -p "${RAPIDS_DOCS_DIR}/cudf/"{html,txt}
+mkdir -p "${RAPIDS_DOCS_DIR}/cudf/html"
 mv build/dirhtml/* "${RAPIDS_DOCS_DIR}/cudf/html"
-mv build/text/* "${RAPIDS_DOCS_DIR}/cudf/txt"
+if [[ "${RAPIDS_BUILD_TYPE}" != "pull-request" ]]; then
+  make text
+  mkdir -p "${RAPIDS_DOCS_DIR}/cudf/txt"
+  mv build/text/* "${RAPIDS_DOCS_DIR}/cudf/txt"
+fi
 popd
 
 rapids-logger "Build dask-cuDF Sphinx docs"
 pushd docs/dask_cudf
 make dirhtml
-make text
-mkdir -p "${RAPIDS_DOCS_DIR}/dask-cudf/"{html,txt}
+mkdir -p "${RAPIDS_DOCS_DIR}/dask-cudf/html"
 mv build/dirhtml/* "${RAPIDS_DOCS_DIR}/dask-cudf/html"
-mv build/text/* "${RAPIDS_DOCS_DIR}/dask-cudf/txt"
+if [[ "${RAPIDS_BUILD_TYPE}" != "pull-request" ]]; then
+  make text
+  mkdir -p "${RAPIDS_DOCS_DIR}/dask-cudf/txt"
+  mv build/text/* "${RAPIDS_DOCS_DIR}/dask-cudf/txt"
+fi
 popd
 
 rapids-upload-docs

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -93,6 +93,7 @@ dependencies:
 - sphinx-autobuild
 - sphinx-copybutton
 - sphinx-markdown-tables
+- sphinx-remove-toctrees
 - sphinxcontrib-websupport
 - streamz
 - sysroot_linux-64==2.17

conda/environments/all_cuda-120_arch-x86_64.yaml

@@ -91,6 +91,7 @@ dependencies:
 - sphinx-autobuild
 - sphinx-copybutton
 - sphinx-markdown-tables
+- sphinx-remove-toctrees
 - sphinxcontrib-websupport
 - streamz
 - sysroot_linux-64==2.17

dependencies.yaml

@@ -472,6 +472,7 @@ dependencies:
           - sphinx-autobuild
           - sphinx-copybutton
           - sphinx-markdown-tables
+          - sphinx-remove-toctrees
           - sphinxcontrib-websupport
   notebooks:
     common:

docs/cudf/source/conf.py

@@ -16,10 +16,12 @@
 # add these directories to sys.path here. If the directory is relative to the
 # documentation root, use os.path.abspath to make it absolute, like shown here.
 #
+import filecmp
 import glob
 import os
 import re
 import sys
+import tempfile
 import xml.etree.ElementTree as ET
 
 from docutils.nodes import Text
@@ -62,13 +64,16 @@ class PseudoLexer(RegexLexer):
     "sphinx.ext.autodoc",
     "sphinx.ext.autosummary",
     "sphinx_copybutton",
+    "sphinx_remove_toctrees",
     "numpydoc",
     "IPython.sphinxext.ipython_console_highlighting",
     "IPython.sphinxext.ipython_directive",
     "PandasCompat",
     "myst_nb",
 ]
 
+remove_from_toctrees = ["user_guide/api_docs/api/*"]
+
 
 # Preprocess doxygen xml for compatibility with latest Breathe
 def clean_definitions(root):
@@ -126,7 +131,13 @@ def clean_all_xml_files(path):
     for fn in glob.glob(os.path.join(path, "*.xml")):
         tree = ET.parse(fn)
         clean_definitions(tree.getroot())
-        tree.write(fn)
+        with tempfile.NamedTemporaryFile() as tmp_fn:
+            tree.write(tmp_fn.name)
+            # Only write files that have actually changed.
+            if not filecmp.cmp(tmp_fn.name, fn):
+                tree.write(fn)
+
+
 
 
 # Breathe Configuration