[Event Hubs] Geo DR - preview

[m-redding]

Contributes to #39367

The concepts being updated here are the following:

• EventData.Offset and EventData.SequenceNumber are now equivalent
• EventData.GlobalOffset now maps to the offset value in the amqp message, but stores it as a string rather than a long

The primary changes in this PR are the following:

• Hide EventData.Offset and add EventData.GlobalOffset, hide old constructor and add new one
• Hide EventData.LastPartitionOffset and add EventData.LastPartitionGlobalOffset, hide old constructor and add new one
• Hide EventPosition.FromOffset and add EventPosition.FromGlobalOffset, hide old constructor and add new one
• Hide LastEnqueuedEventProperties.Offset and add LastEnqueuedEventProperties.GlobalOffset, hide old constructor and add new one
• Add CheckpointPosition.GlobalOffset and new constructor
• Everything in the AMQP layer needs to treat offset as sequence number, and global offset as offset.

The secondary changes are anything that is using the now-hidden properties and methods.

Included are updates to documentation and samples.

[jsquire]

I think this implementation makes sense. I'm fine with ToString telling them about the sequence number when they call this.

[jsquire]

👀

[m-redding]

Yeah I changed my mind about this one from my last PR. I was thinking it just was too confusing to have the constructor just totally ignore a parameter.

I was thinking instead:

• offset -> GetSequenceNumber in EventData, (In theory - will trickle up)
• All references to the old offset properties in our internal implementation are switched to either global offset or sequence number

[jsquire]

I meant the typo that you fixed. I'm reasonably sure that's my bad.

This constructor is used only for testing, I believe, probably because it predates the model factory concept. I can't think of another reason why we needed a public constructor. Am I overlooking anything?

I wonder if we should just confirm that you can create from the model factory and make the new constructor internal?

[m-redding]

Yeah I don't believe it would be used for any other reason. I will update this.

[jsquire]

Since we don't have insight into the service behavior for "I used a sequence number to position a geo-dr namespace reader", I think you made the right call by not adding in the warnings when reading non-geodr checkpoints.

[m-redding]

Technically a breaking change and not ideal - but replication segment is just a placeholder value in the current SDK. Alternatively could just tack global offset on the end and leave replication segment

[jsquire]

I think it's a reasonable change. The type signature is the same. It does break anyone using the name, but since the data was never actually populated, I think it's reasonable and low risk.