Resolved one conflict in pkg/codegen/codegen.go: upstream now uses the
collectComponentTypes / collectOperationTypes / renderBoilerplate helper
decomposition (from PR #2366). Kept that structure and changed the op-
related helper calls to use allOps (paths + webhooks + callbacks)
rather than ops, matching what the previous GenerateTypeDefinitions(t,
spec, allOps, ...) call did, so webhook/callback-derived types continue
to be emitted.

* Add optional type generation for inline schemas

Adds output-options.generate-types-for-anonymous-schemas (default false).
When set, every inline schema that would otherwise generate as an
anonymous Go struct is instead emitted as a named type with a path-
derived name (e.g. GetRolesId200JSONResponseBody_Data). Equivalent to
adding x-go-type-name to every inline schema; when both are present at
the same site, x-go-type-name wins.

Test coverage reorganized: internal/test/anonymous_inner_hoisting/ now
has two subdirectories. The existing always-on hoisting test (oneOf,
anyOf, additionalProperties) moves to implicit/. The new flag-gated
behavior gets its own test in global/, exercising the canonical issue
shape (allOf-merged response with a sibling inline `data` object that
itself contains a nested `role` reference).

Closes: oapi-codegen#1139

* codegen: warn when generate-types-for-anonymous-schemas requires types this config does not emit

The new `output-options.generate-types-for-anonymous-schemas` flag (issue
flow through the same emission path that `generate.models` gates. In a
single-config setup where the flag is set with `models: false` but a
client or server generator is enabled, the generated code emits
references to the hoisted names without declaring them, and `go build`
fails.

Multi-config setups where one config emits `models: true` and a sibling
emits a client/server into the same Go package work correctly today;
the hoisting decisions are deterministic across the two codegen
invocations and the sibling's declarations resolve same-package
references. The failure mode is specifically the single-config case.

The two cases are indistinguishable from a single codegen invocation
(both look like `flag: true, models: false, <ops>: true`), so this
warns rather than errors. Split-config users see an informational
notice they can ignore; misconfigured single-config users get a
heads-up before `go build` complains.

Changes:

- pkg/codegen/configuration.go:
  - Add `GenerateOptions.AnyOperationGenerator()` reporting whether any
    client/server/strict generator is enabled.
  - Add `Configuration.Warnings()`, a cross-field diagnostic that fires
    when `GenerateTypesForAnonymousSchemas && !Models &&
    AnyOperationGenerator()`. Follows the existing
    `GenerateOptions.Warnings()` pattern used for the std-http-server
    Go-version check.
  - Strengthen the doc comment on `GenerateTypesForAnonymousSchemas` to
    call out the single-config failure mode and the multi-config
    consistency requirement explicitly.

- configuration-schema.json: matching description text for the
  IDE-surfaced schema doc.

- cmd/oapi-codegen/oapi-codegen.go: wire `opts.Warnings()` to stderr
  immediately after the existing `Generate.Warnings()` printout, under a
  "cross-field configuration warning(s)" header.

Helper-function cleanup in pkg/codegen/codegen.go:

- Decompose the monolithic `GenerateTypeDefinitions` into
  `collectComponentTypes` (gathers #/components-derived TypeDefinitions),
  `collectOperationTypes` (gathers op-derived TypeDefinitions for the
  boilerplate scan), and `renderBoilerplate` (runs enum / additional-
  properties / union / union+additional passes over the union). The
  emission gating is unchanged — still a single `if opts.Generate.Models`
  block — so this is a pure refactor with no behavioral effect on any
  existing config. Generated output is byte-identical.

No new fixtures or behavioral changes; this commit is intended to be
squashed with the parent `Add optional type generation for inline
schemas` commit before the PR lands.