Update CHANGELOG.md

Author: anfeldma-ms

CHANGELOG.md

@@ -1,42 +1,40 @@
-# Changelog
-
-## 3.3.0
+### 3.3.0
 - Adds a new config option "changefeedstartfromdatetime" that can be used to specify from when the changefeed should be processed. See [Config options](https://github.com/Azure/azure-cosmosdb-spark/wiki/Configuration-references) for details.
 
-## 3.2.0
+### 3.2.0
 - Fixes a regression that caused excessive memory consumption on the executors for large result sets (for example with millions of rows) ultimately resulting in an error "java.lang.OutOfMemoryError: GC overhead limit exceeded"
 
-## 3.1.1
+### 3.1.1
 - Fixes a streaming checkpoint edge case where in the "id" contains "|" character with the "ChangeFeedMaxPagesPerBatch" config applied
 
-## 3.1.0
+### 3.1.0
 - Adds support for bulk updates when using nested partition keys
 - Adds support for Decimal and Float data types during writes to Cosmos DB.
 - Adds support for Timestamp types when they are using Long (unix Epoch) as a value
 
-## 3.0.8
+### 3.0.8
 - Fixes type cast exception when using "WriteThroughputBudget" config.
 
-## 3.0.7
+### 3.0.7
 - Adds error information for bulk failures to exception and log.
 
-## 3.0.6
+### 3.0.6
 - Fixes streaming checkpoint issues.
 
-## 3.0.5
+### 3.0.5
 - Fixes log level of a message left unintentionally with level ERROR to reduce noise
 
-## 3.0.4
+### 3.0.4
 - Fixes a bug in structured streaming during partition splits - possibly resulting in missing some change feed records or seeing Null exceptions for checkpoint writes
 
-## 3.0.3
+### 3.0.3
 - Fixes a bug where a custom schema provided for readStream is ignored
 
-## 3.0.2
+### 3.0.2
 - Fixes regression (unshaded JAR includes all shaded dependencies) which increased build time by 50%
 
-## 3.0.1
+### 3.0.1
 - Fixes a dependency issue causing Direct Transport over TCP to fail with RequestTimeoutException
 
-## 3.0.0
-- Improves connection management and connection pooling to reduce number of metadata calls
+### 3.0.0
+- Improves connection management and connection pooling to reduce number of metadata calls