feat: deep selector hack

[pyAndr3w]

Closes #3336

[pyAndr3w]

Current Selector Structure

The current selector consists of:

• maybe set c3
• if-jmp to internal
• if-jmp to external
• maybe jmp to c3 (or throw 130)

This logic is implemented in the selector-hack get-method.

---

Proposed Optimization

I propose moving the selector into recv_internal (or recv_external, if exists) and making recv_internal (recv_external) the root cell of the contract.

This maneuver will make the bytecode layout of contracts slightly more compact and help avoid unnecessary "jumps" between code cells.

---

Proof of Concept Results

At this stage, proof of concept work for this optimization has been completed. The results are available and they are good.

Current Issues

There is a problem with the absence of the maybe set c3 condition. There is also a problem that tonapi will not be able to properly index contracts with recv_external due to the different location of the function dictionary.

---

Solution

The solution to the problem with the missing maybe set c3 condition is known.

We can obtain a dictionary with all functions at the moment when compilation of recv_internal (recv_external) begins. It will be necessary to manually exclude from the dictionary functions that use the inline flag or are not used in the contract.

For this purpose, Asm.fif has a separate hashmap with function usage flags. However, the recv_internal function is not yet compiled and we don't know which functions are used in it (function flags may change).

To solve this, embedding of fake_* functions has been implemented so that all flags are set at the Asm.fif level and we can safely obtain the necessary dictionary that includes only used functions.

[pyAndr3w]

👉👈 It is also necessary to fix the draft so that it passes absolutely all tests.

[Kaladin13]

Can you please add benchmark table result output here in the conversation? I mean to compare the impact of a new hack

[pyAndr3w]

@Kaladin13

[pyAndr3w]

One test is failing due to the lack of a restriction in the Tact compiler on the minimum method_id.
In the deep selector hack, getters with method_id < 65536 end up in the hashmap with non-inline procedures.

[pyAndr3w]

base jetton e2e tests passed

https://testnet.tonviewer.com/kQBEg_BB-yM5Hyik9O9xuVvVDGEqhKTDwbTVf3rYQrIk_HSN?section=method