Orbit: empty per-analysis conda env / conda not on PATH forces heuristic probing before any local work

_Context: from a controlled experiment in which eight frontier models (opus, sonnet, gpt, gpt2, gemini-2.5-pro, gemini-3.5-flash, deepseek, haiku) each drove an Orbit session to reanalyze the same RNA-seq dataset (Santana 2023 *C. auris*) on 2026-06-02–04. Evidence quotes are from the per-run `activity.jsonl` logs._

## Problem

The per-analysis local environment (`.loom/env`) doesn't exist at session start, ships no bioinformatics packages, and `conda`/`mamba` aren't on `PATH`. So before any local step (DIAMOND RBH, biopython, plotting, PDF) the brain heuristically probes install locations and repeatedly fails -- burning turns and sometimes blocking the run outright.

## Evidence

- **opus**: `conda: command not found`, then probed eight install paths (`for p in $HOME/miniconda3/bin/mamba ...`) before finding one.
- **deepseek**: `EnvironmentLocationNotFound: Not a conda environment`; the env wasn't created until well into the run (18:06:50).
- **gpt2**: conda/mamba unavailable in the shell; fell back to a manual build and `parasail` failed to compile.
- **gpt (1st)**: missing `pdftotext`, `PyPDF2`, `bioblend`, `requests` each blocked progress at different points.

## Impact

Medium-high -- hit 5/8 runs. Every local-compute step is gated behind ad-hoc environment discovery; at best it wastes turns, at worst it blocks the analysis (no DIAMOND → no protein RBH → the gene-ID mapping in #220 can't run).

## Proposed direction

A documented `setup_local_python` affordance (skill or harness step) that probes the common conda/mamba locations once, creates `.loom/env` if absent, and pre-declares the recurring stack (DIAMOND, biopython, matplotlib, adjustText). Pairs naturally with a portable HTML→PDF path (#216) and the macOS coreutils gaps (#218).

## Related

#216 (no PDF affordance), #218 (macOS userland assumptions), #220 (cross-assembly gene-ID skill -- assumes DIAMOND is installable).