perf: fix algorithmic regressions and expand performance-expert agent

[sergio-sisternes-epam]

Description

Fix four high-impact algorithmic performance issues found during a full-codebase
audit, add scaling-guard benchmarks, and expand the performance-expert agent
with a Big O analysis lens so the review panel catches future regressions.

The audit identified 15 issues across the codebase; the top four by impact were:

• O(n) linear scan in has_dependency() called per-package during resolution
• Unconditional full directory scan in HTTP cache on every store() call
• Triple sequential iteration over git refs in marketplace ref resolution
• Repeated environment variable parsing across host-classification functions

Type of change

• Bug fix
• New feature
• Documentation
• Maintenance / refactor

Testing

• Tested locally
• All existing tests pass (18,204 unit tests, 0 failures)
• Added tests for new functionality (if applicable)

7 new benchmark tests in tests/benchmarks/test_lookup_and_cache_benchmarks.py
validate the optimised paths using scaling-guard assertions (ratio-based
regression detection).

Approach

Performance fixes (commit 1):

• dependency_graph.py: Added _repo_url_index: set[str] to DependencyTree;
  has_dependency() is now O(1) set lookup instead of O(n) linear scan.
• http_cache.py: Added _tracked_size running total with a fast-path skip
  in _enforce_size_cap() -- full os.scandir only runs when tracked size
  exceeds the 100MB cap.
• marketplace/builder.py: Replaced triple O(n) iteration with a single-pass
  dict index (tags_by_name, refs_by_name, branches_by_name) + O(1) lookups.
• utils/github_host.py: Extracted repeated os.environ.get().strip().lower().split()
  into _get_ghes_host(), _get_gitlab_single_host(), _get_gitlab_hosts_list().

Benchmarks (commit 2):

• 7 tests covering all three optimisation areas using the existing @pytest.mark.benchmark
  convention and scaling-guard ratio assertions.

Agent expansion (commit 3):

• Expanded performance-expert.agent.md with an "Algorithmic performance lens"
  section covering Big O analysis, missing indexes, unconditional expensive ops,
  import costs, redundant computation, and parallelism opportunities.
• Created references/algorithmic-patterns.md as a lazy-loaded pattern catalogue
  (6 anti-patterns with code examples and a quantification checklist).
• Broadened the panel activation in apm-review-panel/SKILL.md to fire on
  utils/, marketplace/, compilation/ paths and on diffs introducing loops
  over collections or adding os.scandir/subprocess.run calls.

Spec conformance (OpenAPM v0.1)

• N/A -- this PR does not change OpenAPM-observable behaviour.

[Copilot]

Pull request overview

This PR targets install/resolve hot paths by removing a few algorithmic inefficiencies (O(n) scans and repeated parsing) and adds benchmark coverage plus agent/panel guidance to help prevent similar regressions.

Changes:

• DependencyTree.has_dependency() now uses an internal set index for O(1) lookups.
• HttpCache._enforce_size_cap() adds a tracked-size fast path to avoid unconditional directory scans on every store().
• Marketplace explicit ref resolution builds single-pass lookup dictionaries instead of iterating git refs multiple times; plus new benchmarks and updated review-panel/perf-agent guidance.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
tests/benchmarks/test_lookup_and_cache_benchmarks.py	Adds benchmark-style scaling guards for the optimized lookup/cache/env-parsing paths.
src/apm_cli/utils/github_host.py	Extracts env parsing into helpers to reduce repeated normalization work.
src/apm_cli/marketplace/builder.py	Replaces triple-pass remote-ref scans with single-pass indexing for O(1) lookups.
src/apm_cli/deps/dependency_graph.py	Adds _repo_url_index to make has_dependency() constant-time.
src/apm_cli/cache/http_cache.py	Adds _tracked_size and a fast-path skip in _enforce_size_cap().
.apm/agents/references/algorithmic-patterns.md	New reference catalog of algorithmic performance anti-patterns and scaling-guard guidance.
.apm/agents/performance-expert.agent.md	Expands the perf agent with an explicit Big-O / algorithmic performance review lens.
.agents/skills/apm-review-panel/SKILL.md	Broadens perf-expert routing triggers and adds algorithmic-regression heuristics for activation.