In Malayalam Unicode there is a character called Chillu. (ൺ,ൻ,ർ,ൽ,ൾ) are the chillu in Malayalam.

Now the problem,

according to Unicode encoding these chillu can be created by two ways. The Malayalam Unicode table (enwiki) says

• ൺ -> 0d7a,
• ൻ -> 0d7b,
• ർ -> d7c,
• ൽ ->0d7d,
• ൾ ->0d7e,
• ൿ -> 0d7f

are chillu. These are called atomic chillu.

But there is another way to create a chillu like

ന+്+zwj -> 0d28+0d4d+200D. That is [na ന]+[virāma ്]+[ZWJ].

The last thing is called Zero Width Joiner aka ZWJ (Unicode website).

To resolve this problem there was a fix introduced in Mediawiki called $wgFixMalayalamUnicode. If this set as true then all the chillu created by ZWJ will be replaced by the single character chillu (atomic chillu) on save of a page in mediawiki.

This is happening in Malayalam wikipedia and not happening in Wikidata. So there is a good mix of ZWJ chillu and Atomic chillu in Wikidata.

What may be the solution

Enable the $wgFixMalayalamUnicode in Mediawiki used in Wikidata. (But $wgFixMalayalamUnicode is depricated. The release note of 1.35 says it is true by default.)
Do a simple edit on all pages which uses ZWJ Chillu in Wikidata.

Now Wikidata is not converting ZWJ chillu to Atomic chillu. This may be happening because the default langauge of Mediawiki using in Wikidata is not Malayalam. But we need to set $wgFixMalayalamUnicode to true. So How to resolve this problem?

Also there is a thing called $wgAllUnicodeFixes . But this thing will do the fixes for all languages. If this set to true in Mediawiki used by Wikidata then the problem may be solved. @santhosh can you look into this problem?

The rendering issue can be solved by just using a good quality font and latest version. There are many.

The dual encoding issue is quite complicated and has been debated a lot. Practically, the editors should use an input method that is bug free and following atomic chillu. The already done edits can be fixed using a bot. Any search/sort/processing on Malayalam should be aware of this complex encoding model(not limited to Chillu) and process the content accordingly rather than interpreting as raw byte sequences. I do not think that a configuration or extra processing wthin a single software like MediaWiki can solve the complications.

In T266955#6594686, @santhosh wrote:

I do not think that a configuration or extra processing wthin a single software like MediaWiki can solve the complications.

Actually, this problem is not occurring in Malayalam Wikipedia, where, in my info, the variable/parameter mentioned by Ranjithsiji is set to be true. So, in whichever chillu the user is entering, it will actually be transformed to the one supported by Mediawiki, automatically, once it is enabled here I think.