fix: normalize all send request resource release into sr.done

[tisonkun]

This refers to #1071. This follows up to #1120.

This patch contains two fixes with one refactor unsplittably.

• Refactor - collect sendrequest cleanup logic to sr.done and replace it everywhere release sendrequests.
• Fix 1 - only call txn.endSendOrAckOp at the last chunk finished/failed, this fixes [Bug][Producer]Inaccurate transaction endSendOrAckOp for chunked message #1060.
• Fix 2 - blocking reserving resources in front at internalSendAck. Although it breaks chunked messages can pass the permits limitation one by one under DisableBlockIfQueueFull=false, but the Golang client logic is different from Java one, where Golang's internalSendAsync and internalSend do not run in the same thread. I'd prefer to regard the original manner a possible improvement but blocking require full permits should work well for most scenarios.

Together, this patch fixes #1043.

Does this pull request potentially affect one of the following parts:

If yes was chosen, please highlight the changes

• Dependencies (does it add or upgrade a dependency): (yes / no)
• The public API: (yes / no)
• The schema: (yes / no / don't know)
• The default values of configurations: (yes / no)
• The wire protocol: (yes / no)

Documentation

• Does this pull request introduce a new feature? (yes / no)
• If yes, how is the feature documented? (not applicable / docs / GoDocs / not documented)
• If a feature is not applicable for documentation, explain why?
• If a feature is not documented yet in this PR, please create a followup issue for adding the documentation

[tisonkun]

Here is the final failure test -

2023-10-24T21:48:31.4618725Z time="2023-10-24T21:48:23Z" level=info msg="Created consumer" consumerID=1 name=uyats subscription=my-sub topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4619008Z     transaction_test.go:493: 
2023-10-24T21:48:31.4619674Z         	Error Trace:	/pulsar/pulsar-client-go/pulsar/transaction_test.go:493
2023-10-24T21:48:31.4620886Z         	Error:      	Expected nil, but got: &pulsar.Error{msg:"There are some operations that are not completed after the timeout.: TimeoutError", result:3}
2023-10-24T21:48:31.4621142Z         	Test:       	TestAckChunkMessage
2023-10-24T21:48:31.4622274Z time="2023-10-24T21:48:29Z" level=info msg="Closing consumer=1" consumerID=1 name=uyats subscription=my-sub topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4623394Z time="2023-10-24T21:48:29Z" level=info msg="Closed consumer" consumerID=1 name=uyats subscription=my-sub topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4624662Z time="2023-10-24T21:48:29Z" level=info msg="close consumer, exit reconnect" consumerID=1 name=uyats subscription=my-sub topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4625711Z time="2023-10-24T21:48:29Z" level=info msg="Closing producer" producerID=1 producer_name=test topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4626797Z time="2023-10-24T21:48:29Z" level=info msg="Closed producer" producerID=1 producer_name=test topic="persistent://public/default/my-topic-735233632"
2023-10-24T21:48:31.4627034Z --- FAIL: TestAckChunkMessage (5.83s)

[tisonkun]

We do register in internalSendAsync. This call seems like a workaround we ever added as a HACK 🤣

[tisonkun]

Should merge #1120 first

[tisonkun]

@gunli supposed we can still acquire permits with totalChunk size in DisableBlockIfQueueFull=true for fail fast. IMO it can be reasonable and align with Java client's manner. But the current impl is correct also. So I tend to factor out this attempt to a follow-up PR for investigating.

BlockingIfQueueFull is convenient for users, recover the manner in this patch.

[tisonkun]

indent - sr.callback != nil is always true.

[tisonkun]

This is for Fix 2.

[gunli]

I think it is OK, well done.

[nodece]

LGTM

[tisonkun]

Merging...

Thanks for your review!

Further improvement can be made as described above.