fix(import_phoenix_ast): handle wide-format Phoenix exports

[efosternyarko]

Problem

import_phoenix_ast() failed silently or produced incorrect results when given Phoenix export files in wide format (one row per sample, with drug results stored as column triplets: XX (MIC), XX (Interp), XX (Expert)).

The function was written for long-format Phoenix exports (one row per drug) and had no path for the wide format common in Phoenix XLSX batch exports.

Changes

R/import_pheno.R

• Wide-format detection and pivot — after reading the file, detect wide format by checking for multiple (MIC) / (MOC) columns. When detected, pivot to long format: metadata columns are preserved per row, and each drug triplet becomes one row with standardised drug, mic, and Interp columns.
• use_expertized logic during pivot — expert interpretation is preferred over raw Interp per row when use_expertized = TRUE (default), with per-row fallback to Interp where expert is NA.
• Drop untested rows — rows where both MIC and SIR are absent (drug not run for that sample) are removed after pivoting.
• Sample column auto-detection — added "accession" to the sample ID detection patterns, prioritised before "^isolate$". Phoenix wide exports commonly use "Accession" as the sample identifier; previously "Isolate" (the within-batch isolate number) was matched first, collapsing hundreds of samples to a handful of unique IDs.
• Output cleanup — switched final relocate() to select() so pivot intermediate columns (drug, Interp) are not leaked into the returned data frame.

Testing

Tested on three real Phoenix export files:

File	Format	Rows	Samples	Drugs
NMIC-422 MIC Results 20026-Feb.xlsx	wide (headed)	18,420	677	25
PMIC-84 MIC Results 2026 -Feb .xlsx	wide (headed)	32,818	1,228	18
Phoenix-Antibiogramm-Daten.xls	headerless positional	465	20	37

All three return the standardised 8-column output (id, drug_agent, mic, disk, method, platform, pheno_provided, spp_pheno). The existing headerless positional path is unaffected.

[efosternyarko]

CI failure note

The two failing jobs (ubuntu-latest (R-devel) and macOS-latest (R-release)) both fail during dependency setup (r-lib/actions/setup-r-dependencies@v2), before the package check runs — so neither failure is caused by the code changes in this PR.

• ubuntu-latest (R-devel): package or namespace load failed for 'rlang' — binary compiled against an older R-devel ABI, breaks when the daily R-devel build increments
• macOS-latest (R-release): dependency setup exits with code 1 (likely a source compilation failure of a base dependency)

The most recent push to main (2026-03-23) passed all five jobs including both of these. This is a transient infrastructure issue. Could a maintainer re-run the failed jobs when convenient?