fix: reject intrinsic op variants with immediate when literal argument value exceeds its range

[boblat]

Fixes #709.

• when building IR from AWST, always pick the stack-arg variant and let the IR optimiser promote stack args to immediates when the values actually fit.

• add checks on literal value <= 255 to immediate promotions for gitxnas, itxnas, loads/stores in IR optimiser

• add immediate promotion cases in IR optimiser for gaids, gloads, gloadss, txnas, gtxns, gtxnas, gtxnsa, gtxnsas so the new AWST behavior still produces the compact form when possible.

Note: intrincis simplification optimiser does not run for O0 so O0 stats are up.

[engineering-ci]

Name	Status	O0 bytes	O1 bytes	O2 bytes	O0 ops	O1 ops	O2 ops
box_storage	🟢	-	-	-	+1	-	-
calculator	🟢	-	-	-	+3	-	-
hello_world	🟢	+1	-	-	+1	-	-
local_state/LocalStateContract	🟢	-	-	-	+3	-	-
local_state/LocalStateWithOffsets	🟢	-	-	-	+4	-	-
abi_routing/CustomApproval	🟢	-	-	-	+1	-	-
application	🟢	-	-	-	+1	-	-
arc4_conversions/TestContract	🟢	-	-	-	+4	-	-
arc4_dynamic_arrays	🟢	-	-	-	+4	-	-
arc4_types/Arc4BoolEvalContract	🟢	-	-	-	+1	-	-
arc4_types/Arc4BoolTypeContract	🟢	-	-	-	+1	-	-
arc4_types/Arc4DynamicBytesContract	🟢	-	-	-	+1	-	-
arc4_types/Arc4MutationContract	🟢	-	-	-	+10	-	-
arc4_types/Arc4StringTypesContract	🟢	-	-	-	+8	-	-
arc4_types/Arc4StructsTypeContract	🟢	-	-	-	+1	-	-
arc4_types/Arc4TuplesTypeContract	🟢	-	-	-	+1	-	-
arc4_types/MutableParams2	🟢	-	-	-	+4	-	-
array/Contract	🟢	-	-	-	+2	-	-
array/ImmutableArrayContract	🟢	-	-	-	+22	-	-
array/StaticSizeContract	🟢	-	-	-	+2	-	-
asset	🟢	-	-	-	+4	-	-
bytes_backed_ops	🟢	+2	-	-	+6	-	-
dup2_optimization_bug	🟢	+2	-	-	+2	-	-
edverify	🟢	+2	-	-	+4	-	-
everything	🟢	-	-	-	+2	-	-
fixed_bytes_ops/FixedBytesOps	🟢	-	-	-	+10	-	-
group_side_effects/AppExpectingEffects	🟢	-	-1🟢	-1	-	-1🟢	-1
inner_transactions/MyContract	🟢	-	-	-	+1	-	-
intrinsics/ImmediateVariants	🟢	-	-3🟢	-3	+9	-7🟢	-7
large_box_operations/NestedItemArrayUInt64	🟢	-	-	-	+4	-	-
logic_signature/args_complex	🟢	+21	-	-	+22	-	-
logic_signature/args_complex_no_validation	🟢	+11	-	-	+17	-	-
logic_signature/args_simple	🟢	+3	-	-	+3	-	-
logic_signature/dont_use_this	🟢	+9	-	-	+5	-	-
logic_signature/pre_approved_sale	🟢	-	-	-	-	-11🟢	-11
match/MyContract	🟢	-	-	-	+1	-	-
mutable_native_types/Case1WithTups	🟢	-	-	-	+1	-	-
mutable_native_types/Case2WithImmStruct	🟢	-	-	-	+1	-	-
mutable_native_types/Case3WithStruct	🟢	-	-	-	+1	-	-
mutable_native_types/Contract	🟢	-	-	-	+5	-	-
regression_tests/ExtractOpSelection	🆕	+604	+8	+8	+21	+4	+4
regression_tests/GTxnArrayGroupOpSelection	🆕	+25	+18	+18	+12	+9	+9
regression_tests/GTxnArrayIndexOpSelection	🆕	+25	+18	+18	+12	+9	+9
regression_tests/GTxnOpSelection	🆕	+17	+11	+11	+7	+5	+5
regression_tests/GaidOpSelection	🆕	+16	+10	+10	+7	+5	+5
regression_tests/GloadIOpSelection	🆕	+18	+12	+12	+8	+5	+5
regression_tests/GloadTOpSelection	🆕	+18	+11	+11	+8	+5	+5
regression_tests/ReplaceFoldOOB	🟢	+1	-	-	+1	-	-
regression_tests/ReplaceOpSelection	🆕	+549	+543	+543	+13	+11	+11
regression_tests/SubstringEndOOB	🟢	+2	-	-	+2	-	-
regression_tests/SubstringOpSelection	🆕	+605	+8	+8	+21	+4	+4
regression_tests/TxnArrayOpSelection	🆕	+23	+17	+17	+11	+9	+9
regression_tests/arg_op_selection	🆕	+15	+12	+12	+8	+7	+7
simplish	🟢	-	-	-	+3	-	-
state_mutations	🟢	-	-	-	+2	-	-
too_many_permutations	🟢	+2	-	-	+4	-	-
transaction	🟢	-	-1🟢	-1	-	-7🟢	-7
tuple_support/NestedTuples	🟢	-	-	-	+4	-	-
tuple_support/NestedTuplesStorage	🟢	-	-	-	+4	-	-
typed_abi_call/Logger	🟢	+3	-	-	+9	-	-
typed_itxn_abi_call/Logger	🟢	+3	-	-	+9	-	-
undefined_phi_args	🟢	-	-	-	+2	-	-
native_array	🟢	-	-	-	+13	-	-
Total	🟢	+62	-5🟢	-5	+226	-26🟢	-26

[engineering-ci]

Checking stubs for changes and corresponding version bump in origin/fix/op-selection

Last puyapy release: v5.8.1
This branch stubs version: 3.5.0
Last released stubs version: 3.5.0
Main stubs version: 3.5.0

✅ Stub files unchanged
✅ Stub check passed!

[engineering-ci]

Tests	Skipped	Failures	Errors	Time
1691	3 💤	0 ❌	0 🔥	11m 41s ⏱️

[Argimirodelpozo]

So I think that for any op. that would unconditionally fail with something out of bounds for a certain type (gaid, gload, etc.) we should not let that compile but instead emit an error directly. The argument for the different handling vs. what happened with intrinsic folding is that with intrinsic folding it might not be obvious for a dev. that the folding is happening (something might end up foldable after some other opts.) while in this case it'd be a result of direct user input.
Also in this case we wouldn't have the -O0 vs -O1/-O2 divergent behavior, so I'd say lets be aggresive in those instances and fail compilation.

[daniel-makerx]

So I think that for any op. that would unconditionally fail with something out of bounds for a certain type (gaid, gload, etc.) we should not let that compile but instead emit an error directly.

When you say out of bounds here, are you referring to the protocol limit of 16 for number of transactions in a group? If so is that a bound that could ever realistically be changed at the protocol level?

[Argimirodelpozo]

When you say out of bounds here, are you referring to the protocol limit of 16 for number of transactions in a group? If so is that a bound that could ever realistically be changed at the protocol level?

Yes (the T param. for gaid and gload) and also talking about the scratch limit in gload (the I param bounded by 255).
I'd say both of these are very stable protocol bounds. Not that they absolutely couldn't, but if they were to be changed it'd constitute a major breaking change (and I don't think they have ever changed since their introduction), so IMO they are safe to assume hard constants (unlike something like txn fields that I'd say is more subject to change from time to time e.g. when RejectVersion was introduced recently).