[SPARK-23112][DOC] Update ML migration guide with breaking and behavior changes.

Add breaking changes, as well as update behavior changes, to `2.3` ML migration guide.

## How was this patch tested?

Doc only

Author: Nick Pentreath <[email protected]>

Closes #20421 from MLnick/SPARK-23112-ml-guide.

Author: Nick Pentreath

docs/ml-guide.md

@@ -108,7 +108,13 @@ and the migration guide below will explain all changes between releases.
 
 ### Breaking changes
 
-There are no breaking changes.
+* The class and trait hierarchy for logistic regression model summaries was changed to be cleaner
+and better accommodate the addition of the multi-class summary. This is a breaking change for user
+code that casts a `LogisticRegressionTrainingSummary` to a
+` BinaryLogisticRegressionTrainingSummary`. Users should instead use the `model.binarySummary`
+method. See [SPARK-17139](https://issues.apache.org/jira/browse/SPARK-17139) for more detail 
+(_note_ this is an `Experimental` API). This _does not_ affect the Python `summary` method, which
+will still work correctly for both multinomial and binary cases.
 
 ### Deprecations and changes of behavior
 
@@ -123,8 +129,19 @@ new [`OneHotEncoderEstimator`](ml-features.html#onehotencoderestimator)
 **Changes of behavior**
 
 * [SPARK-21027](https://issues.apache.org/jira/browse/SPARK-21027):
- We are now setting the default parallelism used in `OneVsRest` to be 1 (i.e. serial). In 2.2 and
+ The default parallelism used in `OneVsRest` is now set to 1 (i.e. serial). In `2.2` and
  earlier versions, the level of parallelism was set to the default threadpool size in Scala.
+* [SPARK-22156](https://issues.apache.org/jira/browse/SPARK-22156):
+ The learning rate update for `Word2Vec` was incorrect when `numIterations` was set greater than
+ `1`. This will cause training results to be different between `2.3` and earlier versions.
+* [SPARK-21681](https://issues.apache.org/jira/browse/SPARK-21681):
+ Fixed an edge case bug in multinomial logistic regression that resulted in incorrect coefficients
+ when some features had zero variance.
+* [SPARK-16957](https://issues.apache.org/jira/browse/SPARK-16957):
+ Tree algorithms now use mid-points for split values. This may change results from model training.
+* [SPARK-14657](https://issues.apache.org/jira/browse/SPARK-14657):
+ Fixed an issue where the features generated by `RFormula` without an intercept were inconsistent
+ with the output in R. This may change results from model training in this scenario.
 
 ## Previous Spark versions