Implement DAP-13

[branlwyd]

DAP-13: https://www.ietf.org/archive/id/draft-ietf-ppm-dap-13.html

From DAP-09, this will require (roughly):

• Poll collection jobs with GET instead of POST. (DAP-10, PR)
• Upload reports with POST rather than PUT. (DAP-10, PR)
• Remove support for multi-collection of batches, removing max_batch_query_count & the leader-selected by_batch_id query. (DAP-11, PR, [1])
• Remove support for Poplar1. (without multi-collection of batches, Poplar1 is not so useful; DAP-13: drop support for Poplar1 #3480)
• Remove per-task HPKE configurations. (DAP-12, PR)
• Rename "query type" to "batch mode", "fixed-size" to "leader-selected". (DAP-12, PR)
• Remove max_batch_size. (DAP-12, PR)
• Asynchronous aggregation. (DAP-12, PR, Issue)
• Update collection-related messages to match aggregation messages. (DAP-12, PR)
• Replay protection changes. (DAP-12, PR, [2])
• Re-number PrepareError values. (DAP-12, PR)
• Update to VDAF-13. (DAP-13, PR, [3])
• Update to draft-ietf-ppm-dap-taskprov-01. (Spec)
• Rename "prepare error" to "report error". (DAP-13, PR)
• Rename "upload extension" to "report extension". (DAP-13, PR)
• Implement task start time, switch from task end time to task duration. (DAP-13, PR)
• Verify that we always reject reports outside of the task's validity window. (DAP-13, PR)
• Update embedded messages in Query, BatchSelector, PartialBatchSelector to include length prefix. (DAP-13, PR)
• Implement public report extensions. (DAP-13, PR)
• Bump version tag. (DAP-13, PR)

[1] Do not remove the part_batch_selector field, as it is restored in a later change.
[2] This may not require changes, but we should validate that Janus implements the specified behavior.
[3] This depends on a release of libprio-rs implementing VDAF-13 being available.

[branlwyd]

Replay protection changes

For this portion of the work, I think no changes are needed:

• Janus implements leader-side replay checking by only ever including a given report into a single aggregation job. This is stricter than required by DAP, which requires removing/not including reports already marked when generating aggregation jobs, but does not require the report to be marked until aggregation is complete.
• Janus implements helper-side replay checking by checking/marking replay for a report at time of aggregate initialization (by way of writing a scrubbed client report). This is stricter than required by DAP, which only requires that the check must occur before completing the aggregation job.

N.B. there are a few places where we MUST check replay of a report, where Janus does not do so because we implement the check at an earlier point in the protocol. I don't think we should make functional changes to Janus; we might consider an editorial change to DAP to say something like "Report replay MUST be resolved if it has not already been resolved."

[branlwyd]

Verify that we always reject reports outside of the task's validity window

I verified that we will always reject reports that are outside of the task's valid time window (i.e. task_start to task_end, inclusive).

[branlwyd]

I removed the (optional) indication of unknown extension types by the Leader, filing #3505 instead.

[branlwyd]

I filed #3510 to add support for Prio3MultihotCountVec; it's not required for a compliant DAP-13 implementation, but there is little reason (other than prioritization) not to support it.

[branlwyd]

Once #3564 lands, Janus should be able to interop with any DAP-13 aggregator, whether it is using synchronous or asynchronous aggregation.

We should still implement Helper async aggregation (which is my next task), but we should now be DAP-13 complete. (I say "should be" rather than "are" because there is no other DAP-13 aggregator to test against just yet, so we cannot fully validate our implementation.)