Dev fix resource tags

[migrau]

Fix Nextflow resource tiers (CPU/memory) and error handling based on real trace data

Cleans up conf/base.config so process resource requests actually reflect measured usage, fixing two classes of bugs found while validating a resources_test run: processes silently running with cpus=1 (never parallelizing), and several processes so under-provisioned on memory that they either failed permanently after exhausting retries or wasted time retrying from a guaranteed-to-fail baseline.

Changes:

• Consolidate ad-hoc CPU settings into 4 documented tiers (cpu_low/cpu_medium/cpu_high labels + untouched 1-cpu default) and apply them across ~80 modules that were silently defaulting to cpus=1.
• Bump memory for processes verified via trace peak_rss to be under-provisioned: OMEGA/OMEGAMULTI ESTIMATOR/ESTIMATORGLOBALLOC (peak ~24GB vs old 4GB×attempt ceiling of 16GB), ONCODRIVE3D_RUN (peak ~40GB, had no override at all), FILTERNANOSEQNOISE and MUTDENSITYALL:MUTDENSITY (both were guaranteed to fail/retry on the first attempt).
• Fix ONCODRIVEFMLSNVS cohort-level runs (all_samples, Ismale0/1, Samplelocation*) getting OOM-killed by Slurm (confirmed via .command.err, not a timeout despite matching the 6h wall-time default) — decouple cpus from cpu_high's attempt-scaling so memory retries actually gain headroom per core instead of both scaling in lockstep.
• Add errorStrategy = 'ignore' for DNA2PROTEINMAPPING and a retry-then-ignore strategy for PLOTSATURATION, matching known-flaky behavior instead of requiring per-run overrides in the Seqera launch config.

###Test plan:

• Re-run and confirm ONCODRIVEFMLSNVS, ESTIMATOR*, ONCODRIVE3D_RUN, FILTERNANOSEQNOISE and MUTDENSITYALL:MUTDENSITY complete without FAILED/ABORTED tasks
• Spot-check %cpu in the new trace to confirm no tier is now over-provisioned