Zephyr and FreeRTOS jobs do not require 16GB of RAM to build and
constitute the bulk of our CI (by number) quota usage. By moving them to
smaller instances, with 4GB of RAM, we can use the remaining quota
better.

On the infrastructure side we now have a new auto-scaling pool, with
s3.large.2 instances. Those join the existing s3.xlarge.4 (formerly
known as "heavy") and s3.large.8 (formerly known as "light").

The .workspace job uses the s2.large.8 by default, making a safe (but
expensive) default. Specific jobs deviate as follows:

 - zephyr and freertos jobs use s3.large.2
 - two blueprint jobs with more recipes use s3.xlarge.4

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@huawei.com>