feat: First-class caching (#1104)

Caching is now a core Hamilton feature. You can add caching
by using `Builder().with_cache()` and the cache is accessible
via the `Driver.cache` attribute.

The feature is implemented using lifecycle hooks, but this could
change in Hamilton 2.0.

This also introduces the concept of `MetadataStore` and `ResultStore`
which could be further reused with the Hamilton UI or in 
specific deployment scenarios.

This also includes a bug fix for the `@dataloader` decorator which
had accidentally flipped two tags compared to `@load_from` and `from_.`.
The fix includes changes to `@dataloader`, a related test, and to the
visualization function in `graph.create_graphviz_graph`.

The reference documentation page `caching-logic.rst` includes many
TODOs, quirks, limitations, and roadmap items. Eventually, we will need
to deprecate the other caching mechanisms.

---------

Co-authored-by: zilto <tjean@DESKTOP-V6JDCS2>

.gitignore

@@ -152,3 +152,6 @@ examples/**/hamilton-env
 **.pkl
 **.lance
 **.txn
+
+# hamilton default caching directory
+**/.hamilton_cache/

docs/concepts/builder.rst

@@ -179,6 +179,24 @@ Adds `DataSaver` and `DataLoader` nodes to your dataflow. This allows to visuali
     :align: center
 
 
+with_cache()
+--------------
+
+This enables Hamilton's caching feature, which allows to automatically store intermediary results and reuse them in subsequent executions to skip computations. Learn more in the :doc:`/concepts/caching` section.
+
+.. code-block:: python
+
+    from hamilton import driver
+    import my_dataflow
+
+    dr = (
+        driver.Builder()
+        .with_modules(my_dataflow)
+        .with_cache()
+        .build()
+    )
+
+
 with_adapters()
 ---------------