Add Neva linter for style and harness checks

[emil14]

Context

Several style concerns are valid code but should not become compiler errors:

• one import block per file
• import grouping/sorting
• preferring compact chained connections where it improves scanability
• warning on dense networks that should be extracted into helper components
• requiring useful comments for exported entities/components
• naming checks such as Enumerate returning Enumerated<T>, not unrelated Indexed<T>

These are lint-level concerns: a program may fail lint while still compiling, building, and running.

Desired shape

Add a first-class neva lint command as part of the default toolchain, separate from parser/analyzer/compiler semantics.

The compiler should keep rejecting invalid programs, while the linter reports maintainability/style findings. Tooling may run both together, but the implementation and error semantics should stay separate.

Initial rule candidates

• single import block per file
• import ordering/grouping according to docs/style_guide.md
• exported entities must have a leading purpose comment
• suggest chained connections for simple linear flows
• flag dense networks above roughly 5-6 connections as extraction candidates
• naming consistency for common stdlib result types and component/type pairs

Design notes

• Ship out of the box with the main Neva tooling.
• Keep the rule engine extensible so future users can add project/community rules.
• Consider whether this should share infrastructure with a future test runner/plugin system.
• Look at modern toolchain architectures such as Deno/Bun/Vite/Webpack for command/plugin boundaries, but keep Neva's compiler semantics independent.