[ML] Recognize modules takes a long time to return results

[qn895]

Kibana version:
7.13.3

Elasticsearch version:
N/A

Describe the bug:
Currently, it will take a long time to get results from /api/ml/modules/recognize especially if the index pattern includes indices that are frozen.

Steps to reproduce:

1. Create a data view like apm-*-transaction*,auditbeat-*,filebeat-*,logs-*,packetbeat-*,winlogbeat-* where some indices are frozen
2. Open ML Job Settings in the Security App - Detections page
3. Wait until the spinners are done - which may take a minute or so

Expected behavior:
Jobs should be recognized in less than a minute

Any additional context:
We can set the elasticsearch query to respect the UI setting if it's set to ignore frozen in searches.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[peteharverson]

As per suggestion in #117208 (comment), the following should be added to the manifest query for all the data recognizer modules, in a bool query combining it with the other conditions that are already there:

"must_not": { "terms": { "_tier": [ "data_frozen", "data_cold" ] } }

This also needs to be done for the ML modules in the apache and nginx integration packages.

And the change made in #117208 needs to be reverted.

[peteharverson]

Closing issue as the condition

"must_not": { "terms": { "_tier": [ "data_frozen", "data_cold" ] } }

has now been added to the manifest query for all the necessary modules shipped with the ML plugin in #119635 for 8.0.0, and for the Apache and Nginx integration packages.