Break drivers out into separate crates, clean up some technical debt

[abonander]

Original Text

The goal of this PR is to make it possible to build SQLx without any runtime features enabled, or with just runtime-async-std or runtime-tokio enabled (without a TLS provider). As a side-effect, multiple runtime and TLS features can be enabled simultaneously, in which case Tokio (if a Tokio context is available) and native-tls take precedent over async-std and rustls, respectively.

Along the way, I'm fixing some perf issues re. buffering that's causing us to score very poorly on Diesel's metrics suite.

Preliminary benchmarks show a ~50% improvement on "Trivial query (postgres)" for 10,000 rows, bringing SQLx's time back to the same order of magnitude as Diesel's (7ms vs 3ms).

Amended Text

To support developing our planned SQLx Pro offering (#1616) while being able to reuse as much code as possible, we decided to separate out the database drivers into individual crates and do some other significant refactors that have been waiting in the wings. Since I already had a significant refactor pending in this PR, it made sense to roll these changes into it.

Breaking Change Summary

This is not an exhaustive list of breaking changes for 0.7.0, only those that are part of this PR. The full list will be in the CHANGELOG.md entry as usual.

• All deprecated items have been removed.
• The mssql feature and associated database driver has been deleted from the source tree. It will return as part of our planned SQLx Pro offering as a from-scratch rewrite with extra features (such as TLS) and type integrations that were previously missing.
• The runtime-actix-* features have been deleted. They were previously changed to be aliases of their runtime-tokio-* counterparts for backwards compatibility reasons, but their continued existence is misleading as SQLx has no special knowledge of Actix anymore.
  • To fix, simply replace the runtime-actix-* feature with its runtime-tokio-* equivalent.
• The git2 feature has been removed. This was a requested integration from a while ago that over time made less and less sense to be part of SQLx itself. We have to be careful with the crates we add to our public API as each one introduces yet another semver hazard. The expected replacement is to make #[derive(sqlx::Type)] useful enough that users can write wrapper types for whatever they want to use without SQLx needing to be specifically aware of it.
• The Executor impls for Transaction and PoolConnection have been deleted because they cannot exist in the new crate architecture without rewriting the Executor trait entirely.
  • To fix this breakage, simply add a dereference where an impl Executor is expected, as they both dereference to the inner connection type which will still implement it:
    • &mut transaction -> &mut *transaction
    • &mut connection -> &mut *connection
  • These cannot be blanket impls as it triggers an overflow in the compiler due to the lack of lazy normalization, and
    the driver crates cannot provide their own impls due to the orphan rule.
  • We're expecting to do another major refactor of traits when Rust 1.65 releases with Generic Associated Types (GATs).
    This will mean another major release of SQLx but ideally most API usage will not need to change significantly, if at all.
• The fields of Migrator are now #[doc(hidden)] and semver-exempt; they weren't meant to be public.
• The offline feature has been removed from the sqlx facade crate and is enabled unconditionally as most users are expected to have enabled it anyway and disabling it doesn't seem to appreciably affect compile times.
• The decimal feature has been renamed to rust_decimal to match the crate it actually provides integrations for.
• AnyDriver and AnyConnection now require either sqlx::any::install_drivers() or sqlx::any::install_default_drivers() to be called at some point during the process' lifetime before the first connection is made, as the set of possible drivers is now determined at runtime. This was determined to be the least painful way to provide knowledge of database drivers to Any without them being hardcoded.

[abonander]

@tyrelr I'm getting a weird failure in the SQLite describe tests on this branch: https://github.com/launchbadge/sqlx/actions/runs/4069586336/jobs/7009510856#step:7:1062

[abonander]

Once I get the other checks passing I'll merge this and then we can figure out what's going on.

[tyrelr]

@tyrelr I'm getting a weird failure in the SQLite describe tests on this branch: https://github.com/launchbadge/sqlx/actions/runs/4069586336/jobs/7009510856#step:7:1062

I'm assuming a new cargo.lock lead to a new sqlite binary, which happened to optimize that query in a way that broke the logic? But not a huge deal since (at least on my machine) the rebased PR #2253 fixes it.