Raise ProjectNotFoundError when a project doesn't exist (#302)

[tonylampada]

Description

Fixes #302.

A typo'd or deleted project id in workspace.project() (#302) leaves the user with an error that is just the API's raw JSON, so nobody can tell which project failed or why. A 404 now raises ProjectNotFoundError, which names the workspace and project id and points to Trash and API-key access as the likely causes. Serves O8 (one canonical API shape on every surface): SDK, CLI and API now report a missing project the same way.

• rfapi.get_project: 404 → ProjectNotFoundError(RoboflowError, RuntimeError); other statuses unchanged.
• Roboflow.project() now uses the same adapter. It still raises RuntimeError, so existing except clauses keep working.
• The CLI commands that look up a project (project get, infer, version, model, video) get the clearer message for free. They still exit with code 3.
• The red ICN001 CI predates this PR (unpinned ruff 0.17); one unrelated commit aliases datetime as dt to unblock it.
• The backend returns the same 404 for missing, trashed (deletedAt) and no-access projects, so the message lists all three.

Type of change

• Bug fix (non-breaking)

How has this change been tested?

• New automated tests added (tests/test_project_not_found.py, responses). They fail on main and pass here. Full suite: 1233 OK.
• Evidence: https://gist.github.com/tonylampada/062711136aec84e6214a5bf2f14b349a
• Not run against staging: no staging key available.

Any specific deployment considerations

None. This ships in the next SDK release.

Next step

Standalone fix. Next for O8: give CLI/SDK errors for other 404 lookups (versions, workspaces) the same treatment.

PLAN
touches   SDK project load (rfapi.get_project) → Workspace.project(), Roboflow.project(), CLI project lookups: 404 → ProjectNotFoundError naming ws+project (still RoboflowError; CLI catch blocks unchanged)
touches   Roboflow.project() → was RuntimeError(raw json); new error subclasses RuntimeError
touches   upload → unchanged; bad project id now fails at project(), before any upload
tests     project() 404 raises w/ ws+id · tests/test_project_not_found.py
tests     rf.project("ws/x") 404 same, still RuntimeError · same file
tests     non-404 errors unchanged · same file
review    actionable errors — says what failed + what to check
review    backwards compat — exception hierarchy preserved
verify    python -m unittest → all green; pre-commit + mypy clean

🤖 Generated with Claude Code

[tonylampada]

Deliver checks (dm-brain playbooks/deliver):

• Own review + verify recipe: new tests red on main, green here; full suite 1233 OK; pre-commit + mypy clean
• Evidence gist: https://gist.github.com/tonylampada/062711136aec84e6214a5bf2f14b349a
• PR description per contract (why/objective O8, what, tested, deploy, next, plan block)
• Review round: lieutenant ruled no bot on this repo; one jarbas-review round, findings fixed or answered in the comment above
• Not run against staging: no staging key available (per brief).

[tonylampada]

Review round (jarbas-review, local branch scope, one round):

1. ⚠️ test_missing_project_never_reaches_upload was tautological (.upload is never reached once project() raises) → fixed, removed (07041e5).
2. 🤨 404 also covers a wrong or inaccessible workspace slug, but the message only blamed the project → fixed, message now names the workspace slug too.
3. 🤨 Workspace.project / Roboflow.project docstrings didn't document the raise → fixed.
4. 🤨 Roboflow.project('ws/proj/extra') quietly drops the extra part → won't fix, out of scope: predates this PR, not Python API upload fails silently if targeted project does not exist but was before #302.
5. 🤨 Unrelated ICN001 commit → won't fix, out of scope: the red CI predates this PR (CI installs ruff unpinned; 0.17 flags two untouched files). The alias commit only unblocks CI and can be dropped if main fixes it first.

[tonylampada]

Night-loop drill run; closed by Selma.