add neural network description

CHANGELOG.md

@@ -55,6 +55,7 @@ The most noteworthy change of this release is the update of the container's base
   - Refactored helper methods for sending e-mails with specific `Message-ID` headers and the helpers for retrieving + filtering logs, which together help isolate logs relevant to specific mail when multiple mails have been processed within a single test. ([#3786](https://github.com/docker-mailserver/docker-mailserver/pull/3786))
 - **Rspamd**:
   - The `rewrite_subject` action, is now disabled by default. It has been replaced with the new `SPAM_SUBJECT` environment variable, which implements the functionality via a Sieve script instead in favor of being anti-spam service agnostic ([3820](https://github.com/docker-mailserver/docker-mailserver/pull/3820))
+  - `RSPAMD_NEURAL` was added and is disabled by default. If switched on it wil enable the experimental Rspamd Neural network module to add a layer of analysis to spam detection using neural network technology.  ([3833](https://github.com/docker-mailserver/docker-mailserver/pull/3833))
 
 ### Fixes
 

docs/content/config/environment.md

@@ -451,6 +451,17 @@ Can be used to control the score when the [`HFILTER_HOSTNAME_UNKNOWN` symbol](#r
 
 Default: 6 (which corresponds to the `add_header` action)
 
+
+##### RSPAMD_Neural
+
+Can be used to enable or disable the [Neural network module][rspamd-docs-neural-network]. This is an experimental anti-spam weigh method using three neuaral networks in the configuration added here. As far as we can tell it trains itsself by using other modules to find out what spam is. It will take a while (a week or more) to train its first neural network. The config trains new networks all the time and discards of old networks. 
+Since it is experimental it is switched of by default.
+
+- **0** => Disabled
+- 1 => Enabled
+
+[rspamd-docs-neural-network]: https://www.rspamd.com/doc/modules/neural.html
+
 #### Reports
 
 ##### PFLOGSUMM_TRIGGER