finish performance note

Author: bckohan

doc/source/performance.rst

@@ -26,23 +26,33 @@ The following benchmarks show total module loads, import and runtimes. All times
 The first benchmark is for a minimal project that uses the settings file generated by
 :django-admin:`startproject`. You can see that when :pypi:`rich` is installed several hundred extra
 modules must be imported. This adds a few 10s of milliseconds to the import time. To avoid this,
-uninstall :pypi:`rich`. As of :pypi:`Typer` 0.17, :pypi:`rich` is lazily loaded, but until certain
-improvements are made to :pypi:`Typer`, the import time penalty is still incurred by
-:pypi:`django-typer`.
+uninstall :pypi:`rich`. As of :pypi:`Typer` 0.17, :pypi:`rich` is lazily loaded, but `until certain
+improvements <https://github.com/django-commons/django-typer/issues/170>`_ are made to :pypi:`Typer`
+, the import time penalty is still incurred by :pypi:`django-typer`.
 
-.. raw:: html
-    :file: _static/img/minimal_profile.svg
+.. only:: html
+
+   .. raw:: html
+      :file: _static/img/minimal_profile.svg
+
+.. only:: latex
+
+   .. image:: _static/img/minimal_profile.svg
+      :width: 80%
 
 The second benchmark shows the same stats for the :ref:`polls example code <building_commands>`.
 These benchmarks compare the :class:`~django.core.management.BaseCommand` native implementation
 to the :class:`~django_typer.management.TyperCommand` implementation. They show that the overhead of
 using :pypi:`django-typer` is minimal. On the order of a few 10s of milliseconds. Adding
 :pypi:`rich` increases this overhead slightly, but the impact is still barely noticeable.
 
-.. raw:: html
-    :file: _static/img/polls_profile.svg
 
+.. only:: html
+
+   .. raw:: html
+      :file: _static/img/polls_profile.svg
 
-.. success::
+.. only:: latex
 
-    These benchmarks illustrate that adding
+   .. image:: _static/img/polls_profile.svg
+      :width: 80%