Simplify STIM-to-QIR snapshot tests

[João Boechat (joao-boechat)]

This branch makes Stim-to-QIR snapshot tests smaller and easier to review by removing repeated QIR boilerplate from most expectations while retaining the generated instructions, helper definitions, declarations, resource counts, and noise details that distinguish each case. It also makes emitted QIR deterministic.

QIR emission

• Sorts declarations and helper definitions for deterministic output.
• Places the optional qdk_noise attribute before module flags.

Snapshot infrastructure

• Formats successful compilation output into focused body, definitions, and declarations sections, followed by required qubit and result counts.
• Omits standard entry-point initialization, output recording, runtime declarations, attributes, and module metadata from routine snapshots.
• Preserves noise behavior through a uses_noise marker and the effective noise configuration; compiler error expectations remain unchanged.
• Adds dedicated full-QIR checks for empty, measured, and noisy programs so shared entry-point boilerplate, output calls, runtime declarations, attributes, and module flags remain covered.

[orpuente-MS]

I think the choice of syntax is a bit confusing, since the headers body: and declarations: look like valid QIR tags.

[orpuente-MS]

Looks like we are missing the *_record_output calls in the body. I don't consider that to be boilerplate, but rather something that needs validation in the tests.

[orpuente-MS]

This is a nice simplification to aid visual validation. Some general thoughts on the PR:

First: whatever decision gets taken here, these tests will likely go away after the QIR -> Q# AST transition, which is the very next follow up work. With that being said:

1. Just printing the body of the main function creates an asymmetry between the main definition and other function definitions.
2. The main:, declarations: and definitions: headers can be confused with valid QIR tags by someone who didn't design the test format.
3. I believe the *_record_output calls and the function's return instruction are valuable information to validate in the tests' output.

(This last note is a totally optional suggestion: I think you can achieve 80% of the boilerplate simplification by just omitting the declarations, attributes, and module flags. Definitions need review, so they shouldn't be omitted).