cleanup: bump minigraf floor version once minigraf#287, #297, and #298 are fixed

[adityamukho]

Background

`pyproject.toml` was bumped to `minigraf>=1.2.0` (commit `b38f91f`) as part of the 1.2.0 dependency update.

Three bugs are present in the 1.2.0 Rust engine (affecting both the native Python package and the WASM build):

1. bug: EavtKey and AevtKey missing value bytes causes silent data loss when batching multi-valued facts minigraf#287 — batch transact with same entity+attribute loses all but one fact. Any tool in this server that constructs batch transactions with multiple facts sharing the same entity+attribute pair will produce silently wrong results.

2. bug: literal keyword bound in rule call resolves as entity UUID, silently mismatches keyword values in rule body minigraf#298 — literal keyword bound at rule call site resolves as entity UUID, silently mismatches `Value::Keyword` facts in value-position rule body patterns → empty results.

3. bug: magic sets rewriting rejects intermediate variables when first rule arg is bound by literal keyword minigraf#297 — magic sets rewriting flags intermediate variables as "Unbound variable in rule head" when the first rule arg is bound by a literal keyword, causing errors/panics in recursive rules that call other user-defined rules.

Action required

Once all three upstream bugs are fixed and a patch release ships (e.g., 1.2.1), bump the floor version here to `minigraf>=1.2.1` (or whatever the patched version is) to ensure the fixed build is required.

[adityamukho]

State update — two of the three preconditions are now met, and the floor has moved, but this should stay open.

upstream bug	state	in our floor?
project-minigraf/minigraf#297	closed, fixed in v1.2.1	✅
project-minigraf/minigraf#298	closed, fixed in v1.2.1	✅
project-minigraf/minigraf#287	open, milestone 2.0	❌

pyproject.toml now pins minigraf>=1.2.3 (#254). That was driven by a different upstream fix — project-minigraf/minigraf#304, the same-process handle exclusion behind #251/#253 — but it carries the #297 and #298 fixes with it, so this issue's action is satisfied for two of its three items.

#287 is the reason to keep this open. It is a silent data-loss bug (batch transact with the same entity+attribute keeps only one fact) and it is deferred to minigraf 2.0, because fixing it bumps the file format v7→v8 and the project did not want stranded migration code before 2.0 drops v1–v6. So there is no patch release to bump to; waiting is the only option.

The documented workaround stands in the meantime: split multi-valued same-entity+attribute facts into separate transact calls.

Remaining action here is narrower than originally written: bump the floor again when #287 ships in minigraf 2.0. Worth noting that will not be a drop-in bump — 2.0 changes the file format and drops v1–v6 migration, so it needs its own migration check rather than a one-line version change.