feat(backend/docker): add per-step resource limits and OOM score adjustment

[bclermont]

Summary

Add support for per-step memory, CPU limits and OOM score adjustment in the Docker backend via backend_options in workflow YAML.

Motivation

The Docker backend currently only supports global resource limits via agent environment variables (WOODPECKER_BACKEND_DOCKER_LIMIT_MEM, etc.). There is no way to cap individual heavy steps without affecting all steps globally. This leads to OOM situations where the host server freezes because no step-level ceiling exists.

Changes

• Extended BackendOptions struct with resources.limits.{memory,cpus} and oom_score_adj
• toHostConfig now accepts BackendOptions and merges per-step limits over global agent limits
• Added parseMemory helper supporting k/m/g suffixes
• oom_score_adj restricted to [0, 1000] by design — steps can only be made more killable, never protected from the OOM killer
• JSON schema updated so the linter validates backend_options.docker
• Documentation updated with YAML example and option reference table
• All existing tests pass; new tests added for validation, parsing, and toHostConfig merge logic

Usage

steps:
  heavy-build:
    image: golang:1.22
    commands:
      - go build ./...
    backend_options:
      docker:
        resources:
          limits:
            memory: 1g
            cpus: 1.5
        oom_score_adj: 500

Option	Type	Description
resources.limits.memory	string	Max memory (e.g. 512m, 1g). Overrides WOODPECKER_BACKEND_DOCKER_LIMIT_MEM.
resources.limits.cpus	float	Max CPU cores (e.g. 0.5, 1.5). Converted to NanoCPUs.
oom_score_adj	int (0–1000)	OOM killer priority. 0 = default, 1000 = kill first. Negative values rejected by design.

[6543]

we need checks, golbal conf should be able to be limited but not widen

this allows to overwrite conf.resourceLimit.MemLimit to have a greater limit!

[bclermont]

@6543 should I just change it to never go beyond WOODPECKER_BACKEND_DOCKER_LIMIT_MEM ?

[6543]

yes