feat: add per-call options channel and OperationParams projection SPI

[OmarAlJarrah]

Two hand-written sdk-core runtime primitives: a per-call options channel threaded through the context chain, and a minimal SPI for projecting an operation’s inputs into a request. Both are usable by hand today; no code generation is introduced.

Per-call options (Closes #27)

There was no way to override client behaviour for a single call without mutating the shared client. This adds an immutable options bag carried on every link of the DispatchContext -> RequestContext -> ExchangeContext chain.

• CallOptions — a per-call timeout overlay, a response-validation decision, an optional ad-hoc credential, and an open set of typed extension attributes keyed by CallOption<T>. applyDefaults merges a call-site set onto a client baseline with per-field precedence (the receiver wins, falling back to the defaults); CallOptions.NONE is the identity element.
• CallTimeout — a per-phase (connect/write/read/call) overlay where a null phase means "inherit", with the same applyDefaults overlay semantics. The per-call timeout overlay the issue calls for.
• ResponseValidation — a three-state enum (INHERIT/ENABLED/DISABLED) so the inherit case is a first-class non-null value rather than a nullable boolean, matching the issue’s "non-null with defaults" guidance.
• CallContext now exposes callOptions. DispatchContext mints it once at the head of the chain and each promotion (toRequestContext/toExchangeContext) carries it forward unchanged. It defaults to CallOptions.NONE, so every existing construction site is unaffected and the transport SPIs are untouched.

OperationParams projection (Closes #57)

Thin services need a way to project an operation’s inputs into path/query/headers/body that feeds the context chain. The QueryParams multimap (#28) is not in yet, so query contributions are modeled as flat ordered name/value pairs that migrate mechanically once it lands.

• OperationParams — a narrow SPI with pathParams(), queryParams(), headers(), body(), each defaulting to empty so an implementation overrides only what it contributes.
• PathParam / QueryParam — the raw, unencoded value types the projection returns.
• RequestProjector.project(baseUrl, method, pathTemplate, params) — materializes a Request, substituting {name} path placeholders and appending the query string, and owning percent-encoding (path segments escape /; query names/values are form-encoded with + rewritten to %20).
• RequestProjector.projectInto(dispatch, ...) — projects and promotes a DispatchContext straight into a RequestContext, registering it on the chain’s store slot and carrying the chain’s instrumentation, call key, and call options forward.

Tests

New suites: CallOptionsTest, CallTimeoutTest, ResponseValidationTest, CallOptionsPropagationTest (chain threading), OperationParamsTest, RequestProjectorTest (substitution, encoding, repeated/valueless query params, header/body application, error cases, projectInto promotion).

Gated build (module-scoped)

./gradlew :sdk-core:test :sdk-core:ktlintCheck :sdk-core:detekt :sdk-core:apiCheck --no-daemon

All green. Public API snapshot regenerated with ./gradlew :sdk-core:apiDump --no-daemon and committed.

[OmarAlJarrah]

This adds a per-call options channel (CallOptions: per-phase timeout overlay, tri-state response validation, ad-hoc credential, typed attribute map) threaded through the CallContext -> DispatchContext -> RequestContext -> ExchangeContext chain, plus an operation package (OperationParams SPI with PathParam/QueryParam and a RequestProjector that builds a Request from a path template with owned percent-encoding). The context constructors take a trailing callOptions defaulted to CallOptions.NONE, and the API snapshot is regenerated to match. Looks good to merge; just one minor note below.

Issues

Unresolved-placeholder check can miss a malformed segment containing a slash — sdk-core/src/main/kotlin/org/dexpace/sdk/core/operation/RequestProjector.kt:117-123. The guard only inspects the first {...} pair and treats any pair containing a / as not-a-placeholder, so a template like users/{id/extra} with no matching param passes through silently and produces a malformed segment instead of failing fast. The common cases (missing param, extra param) are correctly rejected and tested, and percent-encoding escapes braces in substituted values so a real value can't forge a placeholder, so impact is low. Worth either tightening the check or calling out the slash assumption at the public API boundary so callers know malformed-with-slash templates aren't validated.