1.  Not all types of drawings/diagrams can be imported (as is documented in the UI for doing the import).  If you think it's a diagram that should be able to be imported, you would need to file a defect and attach the word document (or part of it) that  cannot be imported.

2.  Images, diagrams, drawings and the like need to be stored as separate artifacts (import converts the document to what is essentially HTML and HTML does not support "embedded" images).  If you'd like more control over where the images are put as part of the import (e.g., you want to specify the folder the images are to be put in, or the name of the folder), I'd open an enhancement request.

3.  Editing and then saving any rich text document may result in inconsequential (non-visible) changes to be made to the HTML markup (usually made by the browser) or may happen when the markup is re-normalized after the browser makes it messy. This can include things like simply reordering CSS properties attached to text (where there is no visible change of any sort).  This is a side-effect of using browser-based HTML editing.  We are investigating improved compare functions that hopefully will reduce the number of changes that are shown to only those that are of consequence.

4.  There is no relationship between the original (uncoverted) document and the converted document.  So deleting the unconverted one will not affect the converted one (or any of its parts).

5.  Whether you choose to upload an artifact and then later convert it or you chose to import it directly (which effectively is just a combined upload and convert) is really choosing which workflow works better for you.  The two are functionally identical.  Some people prefer uploading (and saving off as a backup) and then converting.  Some prefer the ability of being able to "bulk upload".  Others prefer the simplicity of doing an import directly without any intermediary objects.