Not a direct answer to what you asked, but...
If this is your own work then fixing mistakes directly makes sense,
Whether Calibre can sync in both directions is a question for Calibre users.
AbdSync.exe (in the sig) can sync in either direction to copy the latest.
How well that would work in a Calibre structured system, I don't know.
If this were a Gutenberg book a bookmark and possibly a short note might be more helpful.
Bookmarks are (usually?) hierarchal in the DOM and not the line number of the source HTML.
An RMSDK based device saves bookmarks in XML in a .annot file.
You could write a utility to take the ./annot file and make a snippet out of the EPUB text to email the error to the Gutenberg folks.
Edit: I have another idea.
If your reader saves its place in somewhere accessible, you could do some things without modifying your reader or even making a bookmark.
You read a book, you find an error, you close the book. You open an app and it will be at the same location as you last left the book displaying the raw HTML or the formatted text.
Some people might want to modify and save the file right there.
Others will simply want to snip some text, add a note and append it to the errata that they will eventually send to Gutenberg (or whoever).
More edit: Locations (fragments) compliant with the latest spec are "epubcfi", EPUB Canonical Fragment Identifiers.
The specification:
http://idpf.org/epub/linking/cfi/epub-cfi.html
They look like "epubcfi(/6/4[chap01ref]!/4[body01]/10[para05]/3:5"
Older apps may use simpler pre-spec stuff like "point(/1/4/1/76/1:403)"
Both of these use even numbers for child nodes, odd numbers for content between nodes.
You can see that point uses gratuitous "1" where epubcfi doesn't.
So, with that point example, 4/2-1=1 so go to the (zero-based) nth child of the root,
then go to the 76/2-1=37 nth child of that, get the content of that and 403 chars forward.
Which one does your reader use?