Update example project tests for Tenzir v6

[mavam]

🔍 Problem

example-project is our reference for how a tenzir-test project looks, but
its tests had drifted away from the TQL that Tenzir v6 accepts. Ten of them
failed on a current binary, so anyone following the README saw a broken
project and could not tell harness bugs from example rot.

CI never runs these tests — ci.yml only checks that tenzir-test --help
works on the built wheel — which is why this went unnoticed.

🛠️ Solution

• from_file on an .ndjson file needs an explicit { read_ndjson } parser
  subpipeline.
• The http operator is gone. Sending each event as a request body is now
  each { from_http url, body=$this }, which also applies to the Executor
  example in request.py.
• Module-style names for built-in operators are deprecated, since modules are
  reserved for packages: pipeline::detach, context::create_lookup_table,
  context::update, and context::inspect lose their namespaces.

Baselines are unchanged, so the repaired pipelines produce the same output as
the ones they replace.

💬 Review

Before: 10 failures. After: uvx tenzir-test --root example-project -j 4
leaves only tests that are designed to fail, and they should stay that way:

• tests/fail.tql and tests/shell/exit-code-test.sh demonstrate failure
  rendering, with hand-doctored baselines and exit 42.
• tests/python/pure-python/flaky_coin.py is a coin flip with retry: 5, so
  it fails roughly 1 run in 32 by construction.

That mix is the real usability problem here: a green run is impossible, so
drift like this hides in plain sight. Two follow-ups worth discussing, both
out of scope for this PR:

1. Isolate the intentional failures (own directory, or an expect-failure
   marker) so the rest of the project can run green.
2. Add a CI job that runs example-project against a released tenzir, which
   would have caught every change here.

_{📎 Related: #60}