feat: new text embedding for sparse vector

[cutecutecat]

Part of #459

• Changed svector text representation from "[0, 1, 0, 3, 4]" to pgvector-style "{2:1, 4:3, 5:4}/5"
• Upgraded toolchain for new feature proc_macro_byte_character from upstream

Reminder

The index is from 1 instead of 0 at pgvector

Old	New
[0, 1, 2, 0, 0]	{1:1, 2:2}/5
[0, 1, 2, 3, 4, 5, 6, 7]	{1:1, 2:2, 3:3, 4:4, 5:5, 6:6, 7:7}/8
[5, 6, 7]	{0:5, 1:6, 2:7}/3
[0, 0, 0, 0]	{}/4

[cutecutecat]

The failed CI is due to an upstream uncompatiablity:

• proc-macro2 v1.0.80 introduced proc_macro_byte_character feature, which is recently stablized but not released until 2024-04-12

[usamoi]

The reason why CI fails is that sqllogictest-bin released on crates.io but not on GitHub a few days before (so cargo-binstall fallbacks to build natively). We do not have to update the toolchain.

[usamoi]

I feel not good about indexing from 1. It's not consistent with subscripting.

[VoVAllen]

Let's hold this PR for now, due to the conflict between 1-based array and 0-based array

[VoVAllen]

This is used to support bm25 extension. It can produce string instead of depending on pgvecto.rs/pgvector. cc @cutecutecat

[usamoi]

Parsing must accept all valid inputs and reject all invalid inputs. It should be better if it can be written trivially.

[usamoi]

What about {}/1/2?

[cutecutecat]

What about {}/1/2?

Fixed.

[usamoi]

{1,2,3,4}/5 gives incorrect message:

    OutOfBound {
        dims: 5,
        index: 4294967295,
    },

[usamoi]

{1:1,2:2,}/5 tailing commas are not accepted, but it should be

[usamoi]

Since you have defined ParseState, you can write the code in automata style.