Best approach to split datasets and reports

Hello there, 

I recently started working for a client, and the current top priority is to define the strategy to adopt regarding the distribution of datasets, reports and workspaces inside of the Power BI Service (they are using a Premium capacity).

Basically, this client deals with data from various thematics (calls, sales, maintenance interventions, etc.) that is used to feed different reports.
For the moment, things are not done the correct way, since every report has its own dataset, even though many of them could share the very same dataset, thus avoiding unnecessary redundancy.

One important point to note is that it is exepected that some reports will use data coming from different thematics, for example crossing sales data with calls data.

For the moment, I am thinking about creating a single workspace containing all the datasets, with each dataset dealing with its very own thematic (one dataset for Sales, another one for Calls, etc.), and then create one workspace per thematic (with only reports in it). These workspaces would use shared datasets from the "dataset workspace".

Then, whenever a report would need to cross data coming from different thematic, we would use the "DirectQuery for Power BI datasets and Azure Analysis Services" functionality to handle this need. I don't know yet about how frequent crossing dataset would be needed, but from what I heard it should be quite often.

The reason I am proposing to isolate each thematic in specific datasets is to avoid data redundancy.
I believe that the DirectQuery feature would go along pretty well with this idea whenever it's required, but I am not sure about the restrictions we could face using this method. Especially (but not only) regarding RLS and paginated reports that would need to use two datasets at the same time.
Also, the reason I am not quite sure about this method is that I don't have much experience with DirectQuery over datasets, so I am not fully aware of the restrictions I would be facing later, and I feel like using this feature that way is not why it was developed for in the first place.

The other option I am thinking of is to have one dataset per thematic, plus extra datasets crossing many thematics (ie : "Dataset Sales & Calls"). This option would make things a bit easier maybe, at the cost of the higher number of datasets, and data redundancy (even though we are planning to be using dataflows).

So finally, I have two main questions :

1) Is creating a single workspace for all datasets the right approach ?
Or should I use a "one thematic = 2 workspaces" logic (ie : one workspace for "Sales dataset", one workspace for "Sales reports") ?

2) Is relying on the DirectQuery functionality on datasets to cross-reference them a safe bet ?
Or are there big limitations that make this approach a bit risky or complicated ?
Because for me, the main point of this "one dataset per thematic" approach is to avoid having common tables several times in several datasets (but maybe it's not that much of a problem, especially by using dataflows).
In case this is not the right approach, what would you recommend ? Having more datasets with more tables, even if it means having table "redundancy" between models ?
Or even having one huge dataset with all tables used by all reports ? 

Thank you for your feedback, and do not hesitate to tell me if my totally wrong in my way of thinking !