flowctl: enable self-service task logs

* Update local-stack certificate generation to create certs that work
  with rustls, when used with:

  SSL_CERT_FILE=~/estuary/data-plane-gateway/local-tls-cert.pem

* When reading ops logs, obtain a DPG token which is authorized for the
  task name, rather than the ops collection.

* Drop stats for now to remove flowctl surface area.

* Remove deprecated include-partition / exclude-partition (began to
  panic with auth_prefixes addition to ReadArgs for unknown reasons).

Fix ubuntu references in workflow yaml

journal-client: fix blocking reads of journals

Blocking reads of journals were broken, and would incorrectly return `None`
after reaching the end of a journal.  This manifested as a panic in `flowctl
preview`, which explicitly does not handle the `None` case.  This updates the
`continue_after_eof` function so that we send another read request after
reaching EOF when the read is supposed to be blocking.

ops-catalog: add minTxnDuration to two tasks

Update the committed specs to reflect what's already in production.

flowctl: don't require sops unless we actually need it

The `runtime::unseal` module was a little aggressive about checking to make
sure `sops` and `jq` are both on the path.  This would cause errors, even in
cases where the connector configuration wasn't actually sops-enctypted.  This
moves those checks to be performed only after it's determined that the config
needs to be decrypted.

flowctl: when parsing API responses, use RawValue to not re-order pro…

…perties

agent: don't disable linked materialization bindings

This updates the handling of materialization `sourceCapture`s in the
publications handler, to no longer disable materialization bindings. The agent
will now only ever add bindings to linked materializations, and will never
disable them.

examples: use a view in join-inner example

New connector docs (estuary#1118)

Co-authored-by: Talha Faisal

Merged at Dave's request

flowctl: fix draft author regression

There was a bug in the `clear_draft` function that caused the function to fail
if the draft was not already empty.  This was caused by an incorrect type
parameter in the `api_exec_paginated` callsite. The prior type parameter
`Vec<Value>` was incorrect because the return type of `api_exec_paginated` is
`Vec<T>`, thus the actual type that was expected by the response
deserialization logic ended up being `Vec<Vec<Value>>`.

The fix in this case is to go back to using the plain `api_exec` function,
because the behavior of postgrest for paginated delete calls is
under-specified.

While tracking that down, I added some additional logging to `api_exec`, which
seemed useful enough to keep around.

Also included here is a fix for another tiny bug I noticed, where we would
panic immediately upon a deserialization error instead of returning the error
and letting `main` handle it.