Metadata bucket has lot of overhead even after all the timers are fired

Steps to Repro

• Deployed the following function. There were 2050 docs of metada bucket (expected)

  function OnUpdate(doc,meta) {

  var expiry = new Date();

  expiry.setSeconds(expiry.getSeconds() + 5000);

  var context = {docID : meta.id, random_text : "e6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh0R7Aumoe6cZZGHuh07Aumoe6cZZGHuh07Aumoe6cZZGHuh07Aumoe6"};

  createTimer(NDtimerCallback, expiry, meta.id, context);

  }

  function NDtimerCallback(context) {

  dst_bucket[context.docID] = context.random_text;

  }

• Added 20,000 docs to source bucket
• Metadata had 52,000 on metadata bucket after all the timers were created (expected).
• Waited for all the 20,000 timers to be fired.
• Now metadata bucket has 12,568 docs left where as the expected is around 2050.

I left the system for around 12+ hours in the hope they may get cleared, but that doesn't seem to be the case. I think the only way to clear this extra 10,000 docs is by undeploying a function which is not an ideal scenario. This is 10k overhead for 20k timers created.

I have attached a view query results which emits meta.id from metadata bucket. Let me know if any other info is needed.

This could be a dup of MB-30973 as well. However there is no deploy/undeploy loop here which happens in system test.

I will try to add a automated test to validated this.

Logs attached.