geolocation/sdk: bump compute budget for target-scan instructions

[ben-dz]

Resolves: #3595

Summary of Changes

• Rust and Go geolocation SDKs now prepend a SetComputeUnitLimit(1_400_000) instruction to AddTarget, RemoveTarget, and (Rust only) SetResultDestination transactions. These handlers do an O(n) scan over every target onchain; the default 200K CU per-tx limit exhausted around 743 targets, far below the program's MAX_TARGETS=4096 ceiling. Consumers no longer need to think about compute budgets up to that bound.
• Rust SDK uses an exhaustive match over GeolocationInstruction, so adding a new variant forces an explicit CU-budget decision rather than silently falling through to the default.
• Go SDK renames Build{Add,Remove}TargetInstruction → Build{Add,Remove}TargetInstructions and changes the return type from solana.Instruction to []solana.Instruction. The rename surfaces the breaking type change as a compile error rather than a silent slice/single mismatch; callers feed the returned slice into executor.ExecuteTransactions.
• The Go SDK currently has no SetResultDestination builder, so that handler is only bumped on the Rust side.

Diff Breakdown

Category	Files	Lines (+/-)	Net
Core logic	4	+182 / -22	+160
Tests	4	+129 / -68	+61

Mostly SDK plumbing with matching test updates; no scaffolding/config/docs.

Key files (click to expand)

• smartcontract/sdk/rs/src/geolocation/client.rs — adds TARGET_SCAN_COMPUTE_UNIT_LIMIT, exhaustive-match compute_unit_limit_for(...) helper, and prepends the compute-budget instruction inside GeoClient::execute_transaction. Bucket-coverage unit tests for every variant.
• sdk/geolocation/go/add_target.go — BuildAddTargetInstructions returns []solana.Instruction with a compute-budget prefix.
• sdk/geolocation/go/remove_target.go — same change for BuildRemoveTargetInstructions.
• sdk/geolocation/go/constants.go — adds TargetMutationComputeUnitLimit.
• sdk/geolocation/go/{add_target,remove_target,instructions,executor}_test.go — updated for new names/return type, plus regression tests asserting the prefix is present and requests the right limit.

Testing Verification

• Reproduced the original failure mode (exceeded CUs meter at BPF instruction at ~743 targets) on software-devnet in CH7 prior to the fix; the stress run will now scale to MAX_TARGETS=4096.
• New unit tests in both SDKs assert per-variant bucket assignment and that AddTarget/RemoveTarget (and SetResultDestination in Rust) carry the 1.4M-CU prefix; the Rust tests use exhaustive variant lists so a new instruction can't silently slip through.
• Verified add_target_cu_benchmark.rs budget assumption (1.4M CU sufficient up to N=4095) is unchanged; this PR only flips the SDK to actually request that budget.