…r startup

Restructure JanusInProcess to provision its task via a new provision_task() method once its
ephemeral port is known, so JanusInProcessPair builds one task with consistent real endpoints
and provisions both aggregator views.

Doesn't do anything else.

DAP-18 binds the task's TaskConfiguration -- which contains both the leader
and helper endpoints -- into the HPKE AADs for input and aggregate shares.
AggregatorTask previously stored only peer_aggregator_endpoint, so a party
could not reconstruct both endpoints from the task alone.

Add an own_aggregator_endpoint field (and DB column) alongside the peer
endpoint:

- API-provisioned tasks stamp it at creation from the aggregator's configured
  public_dap_url.
- taskprov tasks populate it from the received TaskConfiguration's endpoint for
  this aggregator's role, which taskprov_opt_in previously discarded.

In the process, we move to the janus_messages::Url.

Add the adapter that yields the canonical TaskConfiguration bound into HPKE
AADs, so the upcoming AAD sites can obtain it from the task alone.

For API-provisioned tasks it is synthesized via build_task_configuration,
pairing the task's own and peer endpoints into leader/helper by role.

For taskprov tasks it is the received wire configuration verbatim. Store that
configuration on the task (new nullable taskprov_task_config column, populated
in taskprov_opt_in). Re-synthesizing it from the stored parameters is not
byte-safe: the VdafConfig<->VdafInstance round trip drops DP strategies and
unknown task extensions would be lost, and any divergence breaks decryption.

…rams

Add builder setters (with_task_info, with_min_batch_size, with_batch_config,
with_vdaf_config, with_task_interval) supplying the parameters the client needs
to reconstruct the task's canonical TaskConfiguration for HPKE AADs. The four
non-optional parameters are validated at build(), which fails fast rather than
deferring to the first upload's AAD construction.

ClientParameters::task_configuration() is the single construction path; it is
not yet AAD-load-bearing (wired in the flag day). In-tree builder callers
(tests, integration harness, interop client) are updated to supply the fields.

… params

Add builder setters (with_helper_endpoint, with_task_info, with_min_batch_size,
with_batch_config, with_vdaf_config, with_task_interval) supplying the parameters
the collector needs to reconstruct the task's canonical TaskConfiguration for
aggregate-share AADs. The collector had no helper endpoint before. Required fields
are validated at build(), which also fails fast on invalid endpoint bytes rather
than deferring to the first collection's AAD construction.

Collector::task_configuration() is the single construction path; it is not yet
AAD-load-bearing (wired in the flag day). Collector::new is removed (it bypassed
the setters and so could no longer build). In-tree callers (tests, integration
harness, interop collector, collect tool) are updated to supply the fields.

Source task_info and the task interval from the provisioned task and thread
them into the client (and TaskParameters) so tests build a TaskConfiguration
that matches the aggregators'. Add TaskBuilder task_info()/task_interval()
getters. Values are threaded but not yet cryptographically bound (the AAD
struct changes land separately), so this is behavior-preserving.

Also fix the get_config expectation to the byte-preserving dap_url (no
trailing slash), which follows from persisting the own endpoint verbatim.