ConvApprox/CastApprox; tuple + array support; doc

[dhardy]

This will likely be released as 0.5:

• conv methods now have default implementations over try_conv methods to ease third-party implementations
• Add ConvApprox and CastApprox for approximate conversions without specific rounding; implement for f64 -> f32
• ConvApprox<T> is auto-implemented over Conv<T>
• Support arrays and tuples for Conv and ConvFloat (and therefore Cast and CastFloat); closes Extend to other types: NonZero*, tuples, arrays, ...? #19
• Update documentation
• MSRV is now 1.53.0

Limitations due to lack of specialization (or other conflicting-trait-impl resolution):

• impl<T> Conv<T> for T was removed
• we cannot implement array/tuple support for ConvApprox
• we cannot auto-implement ConvApprox<T> over ConvFloat<T>

Limitation due to unsized-type support:

• array implementations require T: Copy + Default and may not be optimal

Limitations due to code size:

• Conv supports 0-6 element tuples
• ConvFloat only supports 2-tuples