Docs: machine-specific cluster tree + freshness pass

[cailmdaley]

Update: this PR now also carries the README front door and the basic_execution.md MPI run docs, relocated from #737 so that all user-facing docs live here. The candide cluster walkthrough lives in clusters.md (not duplicated in container.md).

---

Audited every narrative docs page against the current code. The install / container / testing / API pages were already fresh (the conda→uv/container work kept them current); staleness concentrated in cluster docs and a handful of content errors. This PR fixes both.

Machine-specific cluster tree

Cluster guidance was scattered and half-invisible: candide lived only inside container.md (and only on the #737 branch), canfar was split across orphaned pages, and none of the canfar/candide pages were in the sidebar at all.

New single clusters.md under a "Running on a cluster" toctree caption:

• The pattern — the shared truths: the container is the unit of execution, bind-mount your clone at the same path, keep SIFs/cache off a quota-limited $HOME.
• candide (SLURM) — sbatch, the candide_{smp,mpi}.sh scripts, the quota-safe pull → submit, partitions, the MPI/PMIx note.
• CANFAR — the current model (canfar_submit_job / canfar_monitor console scripts), with the deep production walkthrough kept in pipeline_canfar.md (linked, and now in the toctree).
• ccin2p3 — honest stub (not yet containerized).

Deleted obsolete pages

canfar.md (old curl-VM submission, superseded by canfar_submit_job), pipeline_v2.0.md (personal paths, a missing script), work_flow_v2.0.md (an unrealized planning wishlist) — all three orphaned. The v2.0 wishlist was preserved in the team's felt store before deletion.

Content fixes

• dependencies.md — rewritten against pyproject.toml: reframed around the abstract-minimums + uv.lock SSOT (was "pinned per release"); ngmix now points at the aguinot/ngmix@stable_version fork (was esheldon upstream); dropped the phantom CDSclient; added the missing CANFAR/data stack (vos, skaha, canfar, cs_util, astroquery, reproject, h5py, numba).
• post_processing.md — dropped the removed rho-statistics step and the dead prepare_tiles_for_final command; legacy banner → sp_validation.
• random_cat.md — legacy banner; fixed random_runner → random_cat_runner.
• pipeline_canfar.md — flagged the matched-star / coverage-mask helpers that moved to sp_validation.
• basic_execution.md — replaced the conda-era "activate the environment" framing with the container reality. MPI sections deferred pending the Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts #737 keep/drop decision.
• Cosmetics: configuration.md (conifg→config, NUMBERING_LIST→NUMBER_LIST), contributing.md (Pleas→Please), module_develop.md (src/shapepipe/modules).

Verification

Local sphinx-book-theme build succeeds. The one new warning the tree introduced (a clusters.md heading anchor) is fixed; remaining warnings are all pre-existing (the autosummary API page needs the installed package; the multiple-toctree notice fires on every page).

Relationship to the other docs PRs

• Versioned docs with a version switcher #738 (versioned docs) is what makes any of this publishable from develop — until it lands, the site still serves a stale master.
• The candide section describes the SLURM scripts from Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts #737; it reads correctly once Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts #737 has merged to develop (on develop today those scripts are still PBS/conda). So this is best merged after Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts #737.

— Claude on behalf of Cail