Alignment of HTML with localized URLs results in TM segments losing the hyperlink tags in target segment

The title is a bit misleading, but I'm not able to express the weirdness in more meaningful way :-\.
Either I'm missing some fundamental concept, or there is something weird happening in Studio Alignment (or in Studio in general)...

I need to align a MadCap Flare project containing ~1200 files in total (1100+ HTMLs and the rest MadCap internal files... all of which normally ARE localizable, but Studio built-in filetype ignores them... probably yet another unfinished work :( ).
It's an online help, i.e. contains zillions of hyperlinks. The hyperlinks are localized, i.e. different in each language.

And I have found that TMs created by aligning such files are somehow 'losing' the hyperlinks in TM lookup window.

The problem can be easily simulated on trivial HTML files:

<html>
<head></head>
<body>
Click <a href="http://www.example.com/en-us/"> to continue.
</body>
</html>
<html>
<head></head>
<body>
Klicken Sie <a href="http://www.example.com/de-de/"> um fortzufahren.
</body>
</html>

Aligning these two files using an empty TM with default settings results in this (no surprise here, all as expected):

Saving the alignment as SDLXLIFF and examining the SDLXLIFF content shows that hyperlink tags are present in both source and target (again, no suprise here... file was re-formatted and some tags folded for better visibility):
The only potential "issue" here are the different IDs of the <g> element, see below...

Importing this SDLXLIFF in the empty TM results in this content of the TM... again showing the different IDs:

And trying to translate the original source file using that TM (and Alignment penalty set to 0, for the information completeness) shows this - the hyperlink tags are completely missing in the TM lookup pane and are NOT inserted in the translation!

Now this is a BIG issue here since the intention is to automatically pre-translate files using the TM automatically (using 0 Alignment penalty, thus e.g. expecting ALL source files originally used for alignment being fully translated from the TM).

What is wrong here?!
I kind of understand the reason for different IDs - the content of the tags is not identical, so they get different IDs - but how am I supposed to approach the task then?

If I let extract the "href" attribute content for translation, I will get huuuuge amount of extra wordcount... and no one is going to pay for that extra "translation"!
Acronyms auto-substitution does nto work for URLs either...