Consolidate example suite + fix four cross-construct codegen bugs

[MelbourneDeveloper]

TLDR

Consolidates the over-isolated examples/tested suite into fewer cross-cutting examples, relocates the language spec under docs/specs, and fixes four LLVM-codegen bugs that the new construct-mixing examples exposed — each pinned by a regression test.

Details

Test consolidation (net −2079 lines across examples). 69 single-purpose .osp examples were folded into omnibus files (feature_omnibus, list_basics, map_basics, comprehensive_math, boolean_consolidated, pattern_matching_complete, string_pipeline, …) plus a handful of new mixed examples. examples_test.go moved its giant inline expected-output map onto sibling .expectedoutput files; the runner still fails loudly when an example has neither a file nor a map entry (no silent coverage loss).

Spec relocation. 17 spec docs moved compiler/spec/ → docs/specs/ (100% renames) with the website spec-copy script and pages updated to match.

New cross-cutting mixup example. examples/tested/effects/fiber_effects.osp now threads fibers + algebraic effects + records + a union + pattern matching + persistent List/Map through one program instead of siloing each construct.

Four codegen bug fixes (compiler/internal/codegen/), each surfaced only when constructs are combined:

• spawn of a non-constant captured arg (record / runtime value) emitted use of undefined value — the fiber closure referenced parent-function SSA values. Fixed by spilling non-constant captures to per-spawn module globals and reloading them inside the closure (fiber_generation.go).
• Discriminated-union construction used a stack alloca and returned the pointer, so a union returned from a function dangled — garbage int payloads, or a segfault when a string payload was dereferenced. Now heap-allocated via malloc with an exact struct size (expression_generation.go).
• m[key] on a runtime-built map (Map()/mapSet) read the opaque OspreyMap* handle as the flat-literal {i64,i8*} array and segfaulted. Now routed through the C runtime get (osprey_map_contains/osprey_map_get), shared with the mapGet builtin (collection_codegen.go, expression_generation.go).

How Do The Automated Tests Prove It Works?

Four regression examples under examples/tested/basics/knownbugs/ pin each bug with its correct expected output; before the fixes they failed with the exact symptoms above (IR error / segfault / wrong output), and now pass through the standard TestBasicsExamples runner:

• TestBasicsExamples/knownbugs/bug1_spawn_record → 27\n27
• TestBasicsExamples/knownbugs/bug2_string_union_payload → done 27\nfailed empty
• TestBasicsExamples/knownbugs/bug3_map_built_index → a=9\nlen 2
• TestBasicsExamples/knownbugs/bug4_union_return_arg → done 27\ndone 27\nfailed 99

The mixup is verified by TestEffectsExamples/fiber_effects (exact stdout match). The full go test ./... for the compiler passes (integration + unit/codegen), and make lint/make build are clean, confirming no regression across the existing union/record/fiber/collection examples.