Add support for PREPARE block in stim

[joao-boechat]

Add support for PREPARE blocks in Stim

Summary

Adds a PREPARE { ... } block to the Stim front end: a repeat-until-success region used to deterministically prepare a state that otherwise only occurs probabilistically. This replaces the previous #!preselect_* comment-annotation mechanism.

How it works

• A PREPARE block runs its body, then re-runs the entire block until every REQUIRE inside it is satisfied.
• REQUIRE rec[-1] rec[-2] ... reads measurement records (indexed backwards from the most recent measurement) and combines them by XOR (parity). If the parity is 1 the block restarts; if 0 execution continues.
• A target can be negated (REQUIRE !rec[-1]), which flips that record's contribution to the parity — i.e. it requires the record to be 1 instead of 0.
• Each block is lowered to a labeled basic block (prepare_N) plus a conditional branch back to it. REQUIRE may only reference records produced inside its own enclosing block; nested blocks are supported.

Modifications

• qir.rs — lowering for PREPARE/REQUIRE: prepare/continue labels, measurement-record resolution with scope tracking, XOR parity fold, negation (xor … , true), and new diagnostics (RequireOutsidePrepareBlock, MeasurementRecordOutOfBounds, MeasurementRecordOutOfScope).
• parser.rs — parse PREPARE blocks and REQUIRE; allow negation on measurement-record targets.
• lex.rs — drop the old #! preselect-annotation lexing.
• Tests — new qir/tests/prepare_block.rs and parser target tests; removed the obsolete qir/tests/preselect.rs.
• Sample — samples/notebooks/stim_prepare.ipynb demonstrating preselection, negated requirements, multiple/parity requirements, and nested blocks.